Network Shaofu. Peng
Internet-Draft ZTE Corporation
Intended status: Standards Track Zongpeng. Du
Expires: 9 February 2025 China Mobile
Kashinath. Basu
Oxford Brookes University
Zuopin. Cheng
New H3C Technologies
Dong. Yang
Beijing Jiaotong University
Chang. Liu
China Unicom
8 August 2024
Deadline Based Deterministic Forwarding
draft-peng-detnet-deadline-based-forwarding-12
Abstract
This document describes a deadline based deterministic forwarding
mechanism to IP/MPLS network, as well as corresponding resource
reservation, admission check, policing, etc, to provide guaranteed
latency bound. Especially, latency compensation with core stateless
is discussed to replace reshaping to be suitable for Diff-Serv
architecture [RFC2475].
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 9 February 2025.
Copyright Notice
Copyright (c) 2024 IETF Trust and the persons identified as the
document authors. All rights reserved.
Peng, et al. Expires 9 February 2025 [Page 1]
Internet-Draft Deadline Queueing Mechanism August 2024
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1. Requirements Language . . . . . . . . . . . . . . . . . . 5
2. EDF Scheduling Overview . . . . . . . . . . . . . . . . . . . 5
2.1. Planned Residence Time of the DetNet Flow . . . . . . . . 6
2.2. Delay Levels Provided by the Network . . . . . . . . . . 7
2.3. Relationship Between Planned Residence Time and Delay
Level . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4. Relationship Between Service Burst Interval and Delay
Level . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3. Sorted Queue . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1. Scheduling Mode for PIFO . . . . . . . . . . . . . . . . 8
3.2. Schedulability Condition for PIFO . . . . . . . . . . . . 8
3.2.1. Conditions for Leaky Bucket Constraint Function . . . 9
3.2.2. Schedulability Condition Analysis for On-time Mode . 11
3.3. Buffer Size Design . . . . . . . . . . . . . . . . . . . 11
4. Rotation Priority Queues . . . . . . . . . . . . . . . . . . 11
4.1. Alternate Queue Allocation Rules . . . . . . . . . . . . 14
4.2. Scheduling Mode for RPQ . . . . . . . . . . . . . . . . . 14
4.3. Schedulability Condition for RPQ . . . . . . . . . . . . 14
4.3.1. Schedulability Condition for Alternate QAR . . . . . 15
4.3.2. Conditions for Leaky Bucket Constraint Function . . . 15
4.3.3. Schedulability Condition Analysis for On-time Mode . 17
4.4. Buffer Size Design . . . . . . . . . . . . . . . . . . . 17
5. Reshaping . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6. Latency Compensation . . . . . . . . . . . . . . . . . . . . 18
6.1. Get Accumulated Residence Time Deviation . . . . . . . . 19
6.2. Get Allowable Queueing Delay . . . . . . . . . . . . . . 19
6.3. Scheduled by Allowable Queueing Delay . . . . . . . . . . 20
7. Option-1: Reshaping plus Sorted Queue . . . . . . . . . . . . 22
8. Option-2: Reshaping plus RPQ . . . . . . . . . . . . . . . . 23
9. Option-3: Latency Compensation plus Sorted Queue . . . . . . 24
9.1. Packet Disorder Considerations . . . . . . . . . . . . . 25
10. Option-4: Latency Compensation plus RPQ . . . . . . . . . . . 27
10.1. Packet Disorder Considerations . . . . . . . . . . . . . 29
11. Jitter Performance by On-time Scheduling . . . . . . . . . . 31
12. Resource Reseravtion . . . . . . . . . . . . . . . . . . . . 34
12.1. Delay Resource Definition . . . . . . . . . . . . . . . 35
Peng, et al. Expires 9 February 2025 [Page 2]
Internet-Draft Deadline Queueing Mechanism August 2024
12.2. Traffic Engineering Path Calculation . . . . . . . . . . 37
13. Policing on the Ingress . . . . . . . . . . . . . . . . . . . 37
14. Overprovision Analysis . . . . . . . . . . . . . . . . . . . 40
15. Compatibility Considerations . . . . . . . . . . . . . . . . 40
16. Deployment Considerations . . . . . . . . . . . . . . . . . . 42
17. Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . 43
17.1. Examples . . . . . . . . . . . . . . . . . . . . . . . . 45
17.1.1. Heavyweight Loading Example . . . . . . . . . . . . 45
17.1.2. Lightweight Loading Example . . . . . . . . . . . . 48
18. Taxonomy Considerations . . . . . . . . . . . . . . . . . . . 53
19. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 54
20. Security Considerations . . . . . . . . . . . . . . . . . . . 54
21. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 54
22. References . . . . . . . . . . . . . . . . . . . . . . . . . 54
22.1. Normative References . . . . . . . . . . . . . . . . . . 54
22.2. Informative References . . . . . . . . . . . . . . . . . 57
Appendix A. Proof of Schedulability Condition for RPQ . . . . . 58
Appendix B. Proof of Schedulability Condition for Alternate QAR of
RPQ . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 61
1. Introduction
[RFC8655] describes the architecture of deterministic network and
defines the QoS goals of deterministic forwarding: Minimum and
maximum end-to-end latency from source to destination, timely
delivery, and bounded jitter (packet delay variation); packet loss
ratio under various assumptions as to the operational states of the
nodes and links; an upper bound on out-of-order packet delivery. In
order to achieve these goals, deterministic networks use resource
reservation, explicit routing, service protection and other means.
Resource reservation provides dedicated resources (such as bandwidth,
buffer space, time slots, etc.) to DetNet flows. Explicit routing
ensures the stability of the route and does not change with the real-
time change of network topology. Service protection reduces the
packet loss by sending multiple DetNet flows along multiple disjoint
paths at the same time.
[P802.1DC] described some Quality of Service (QoS) features specified
in IEEE Std 802.1Q, such as per-stream filtering and policing,
queuing, transmission selection, stream control and preemption, in a
network system which is not a bridge. The internal structure of IP/
MPLS routers may also be based on these components to describe the
scheduling process of packets. In the presence of admission check,
policing, reshaping, a large number of packet scheduling techniques
can provide bounded latency. However, many solutions may result in
an inefficient use of network resources, or provide an overestimated
latency. Currently the underlying scheduling mechanisms in IP/MPLS
Peng, et al. Expires 9 February 2025 [Page 3]
Internet-Draft Deadline Queueing Mechanism August 2024
networks generally use SP (Strict Priority) and WFQ (Weighted Fair
Queuing), and manage a small number of priority based queues. They
are rate based schedulers.
For SP, the highest priority queue can consume the total port
bandwidth, while for WFQ scheduler, each queue may be configured with
a pre-set rate limit. Both of them can provide the worst-case
latency, but evaluation is generally overestimated. In the case
where the network core supports reshaping per flow (or optimized
reshaping as provided by [IR-Theory]), the worst-case latency of a
flow is approximately equal to the aggregated burst of the traffic
class divided by the rate limit of that traffic class (note that a
rate-based scheduler may refer to [Net-Calculus] to obtain its rate-
latency service curve and get a more tighter evaluation). When the
network core does not implement reshaping, multiple flows sharing the
same priority may form burst cascade, making it more difficult or
even impossible to evaluate the worst-case latency of a single flow.
[EF-FIFO] discusses the SP scheduling behavior in this core-stateless
situation, which requires the overall network utilization level to be
limited to a small portion of its link capacity in order to provide
an appropriate bounded latency.
To address the overestimation issue of rate based scheduling (i.e.,
if want a low latency, may be forced to allocate a large service
rate.), according to [EDF-algorithm], an EDF (earliest-deadline-
first) scheduler, which always selects the packet with the shortest
deadline for transmission, is an optimal scheduler for a bounded
delay service in the sense that it can support the delay bounds for
any set of connections that can be supported by some other scheduling
method. EDF is a delay-based scheduler, which further distinguishes
traffic in terms of time urgency, rather than rough traffic classes.
This document introduces EDF scheduling mechanism to IP/MPLS network,
as well as corresponding resource reservation, admission control,
policing, etc, to provide guaranteed latency, as a supplement to IEEE
802.1 TSN mechanisms (please refer to
[I-D.hp-detnet-tsn-queuing-mechanisms-evaluation] for their
challenges meeting large scaling requirements). It is a layer-3
solution and can operate over different types of QoS sensitive layer
2 including TSN but is not an alternative to TSN. Especially, a
latency compensation based option is recommonded to replace reshaping
to be suitable for Diff-Serv architecture [RFC2475]. This document
also discusses two scheduling behaviors: in-time scheduling and on-
time scheduling. The former only provide bounded delay, while the
latter further provide bounded jitter.
Peng, et al. Expires 9 February 2025 [Page 4]
Internet-Draft Deadline Queueing Mechanism August 2024
1.1. Requirements Language
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
2. EDF Scheduling Overview
The EDF scheduler assigns a deadline for each incoming packet, which
is equal to the time the packet arrives at the node plus the latency
limit, i.e., planned residence time (D), see Section 2.1. The EDF
scheduling algorithm always selects the packet with the earliest
deadline for transmission.
The precondition for EDF to work properly is that any DetNet flow
must always satisfy the given traffic constraint function when it
reaches a certain EDF scheduler. Therefore, it should generally
implement traffic regulation at the network entrance to ensure that
the admitted traffic complies with the constraints; And, implement
reshaping on each intermediate node to temporarily cache packets to
ensure that packets entering the EDF scheduler queue comply with the
constraints. However, reshaping per flow is a challenge in large-
scaling networks. Some core stateless optimization method need to be
considered.
Another challenge of EDF scheduling is that queued packets must be
sorted and stored according to their deadline, and whenever a new
packet arrives at the scheduler, it needs to perform search and
insert operations on the corresponding data structure, e.g., a List,
a PIFO (put-in first-out) queue, or other type of sorted queue, at
line rate. [RPQ] described rotating-priority-queues that approximate
EDF scheduling behavior, and do not require deadline based sorting of
queued packets, simplifying enqueueing operations.
According to the above two challenges and the potential optimization
methods, we will obtain four combination solutions. Operators should
choose appropriate solutions based on the actual network situation.
This document suggests using option-3 or option-4, which are referred
to as CEDF (latency Compensation EDF). CEDF adjusts and sorts the
arrival flows through the latency compensation factor carried in the
packets, ensuring that the flows arrived at the EDF scheduler always
conform to their constraints, avoiding the network core from
maintaining the flow states to meet large scaling requirements.
* option-1: Reshaping plus sorted queue.
Peng, et al. Expires 9 February 2025 [Page 5]
Internet-Draft Deadline Queueing Mechanism August 2024
* option-2: Reshaping plus RPQ.
* option-3: Latency Compensation plus sorted queue.
* option-4: Latency Compensation plus RPQ.
2.1. Planned Residence Time of the DetNet Flow
The planned residence time (termed as D) of the packet is an offset
time, which is based on the arrival time of the packet and represents
the maximum time allowed for the packet to stay inside the node.
For a deterministic path based on deadline scheduling, the path has
deterministic end-to-end delay requirements. The end-to-end delay
includes two parts, the accumulated residence time and the
accumulated link propagation delay. Subtract the accumulated link
propagation delay from end-to-end delay budget to obtain the
accumulated residence time. The accumulated residence time may be
shared equally by each node along the path to obtain the average
planned residence time of each node, or each node may have different
planned residence time. Note that the link propagation delay in
reality may be not always constant, e.g., due to the affection of
temperature, we assume that the tool for detecting the link
propagation delay can sense the changes beyond the preset threshold
and trigger the recalculation of the deterministic path.
There are many ways to indicate the planned residence time of the
packet.
* Carried in the packet. The ingress PE node, when encapsulating
DetNet flows, can explicitly insert the planned residence time
into the packet according to SLA. The transit node, after
receiving the packet, can directly obtain the planned residence
time from the packet. Generally, only a single planned residence
time needs to be carried in the packet, which is applicable to all
nodes along the path; Or insert a stack composed of multiple
planned residence time, one for each node.
[I-D.peng-6man-deadline-option] defined a method to carry the
shared planned residence time in the IPv6 packets.
[I-D.pb-6man-deterministic-crh], [I-D.p-6man-deterministic-eh]
defined methods to carry the stack of planned residence time in
the IPv6 packets.
* Included in the matched local FIB entry or policy entry. An
implementation should support the policy to forcibly override the
planned residence time obtained from the packet.
Peng, et al. Expires 9 February 2025 [Page 6]
Internet-Draft Deadline Queueing Mechanism August 2024
2.2. Delay Levels Provided by the Network
The network may provide multiple delay levels on the outgoing port,
each with its own delay resource pool. For example, some typical
delay levels may be 10us, 20us, 30us, etc.
In theory, any additional delay level can be added dynamically, as
long as the buffer and remaining bandwidth on the data plane allow.
The quantification of delay resource pool for each delay level is
actually based on the schedulability conditions of EDF. This
document introduces two types of resources per delay level:
* Burst: represents the amount of bits bound that a delay level
provided.
* Bandwidth: represents the amount of bandwidth bound that a delay
level provided. The bandwidth possessed by a certain delay level
is also known as the service rate of that delay level.
For more information on the construction of resource pools, please
refer to Section 3.2 and Section 4.3.
2.3. Relationship Between Planned Residence Time and Delay Level
The planned residence time (D) is the per-hop latency requirement of
the flow, while the delay level (d) is the capability provided by the
link.
Generally, we only need to design a limited number of delay levels to
support a larger number of per-hop latency requirement. For example,
there are delay levels such as d_1, d_2, ..., and d_n, In the
resource management of the control plane, we assign d_i resources to
all D that meet d_i <= D < d_(i+1).
2.4. Relationship Between Service Burst Interval and Delay Level
Although we generally prefer to have the service burst interval (SBI)
greater than the maximum delay level, there is actually no necessary
association between SBI and delay level.
Peng, et al. Expires 9 February 2025 [Page 7]
Internet-Draft Deadline Queueing Mechanism August 2024
Firstly, can a flow with small SBI (such as 10us) request a larger
delay level (such as 100us)? Yes. It seems that during a longer
residence time caused by delay level, there will be multiple rounds
of burst interval packets leading to bursts accumulation. However,
these packets can be distinguished and sent in sequence. In fact, we
can multiply the original SBI by several times to obtain the expanded
SBI (which includes multiple original bursts), with a length greater
than the requested delay level, to get the preferred paradigm.
Secondly, can a flow with large SBI (such as 1ms) request a smaller
delay level (such as 10us)? This is certainly yes.
3. Sorted Queue
[PIFO] defined the push-in first-out queue (PIFO), which is a
priority queue that maintains the scheduling order or time. A PIFO
allows elements to be pushed into an arbitrary position based on an
element's rank (the scheduling order or time), but always dequeues
elements from the head.
3.1. Scheduling Mode for PIFO
A PIFO queue may be configured as either in-time or on-time
scheduling mode, but cannot support both modes simultaneously.
In the in-time scheduling mode, as long as the queue is not empty,
packets always departured from the head of queue (HoQ) for
transmission. The actual bandwidth consumed by the scheduler may
exceed its preset service rate C.
In the on-time scheduling mode, if the queue is not empty and the
rank of the HoQ packet is equal to or earlier than the current system
time, the HoQ packet will be sent. Otherwise, not.
3.2. Schedulability Condition for PIFO
[RPQ] has given the schedulability condition for classic EDF that
based on any type of sorted queue with in-time scheduling mode.
Suppose for any delay level d_i, the corresponding accumulated
constraint function is A_i(t). Let d_i < d_(i+1), then the
schedulability condition is:
sum{A_i(t-d_i) for all i} <= C*t (Equation-1)
where, C is service rate of the EDF scheduler.
Peng, et al. Expires 9 February 2025 [Page 8]
Internet-Draft Deadline Queueing Mechanism August 2024
It should be noted that for a delay level d_i, its residence time is
actually contributed by its own flows and all other more urgent delay
levels. Based on the schedulability conditions, we can choose the
traffic arrival constraint function according to the preset delay
level, or we can choose the delay level according to the preset
traffic arrival constraint function.
When setting up new flows in the network, admission check based on
schedulability condition must be excecuted on each link that the flow
passes through.
Here, A_i(t) defines the upper limit of eligible arrivals of delay
level d_i, and should not be treated as the actual arrivals (we mark
it as a_i(t) for distinction). As described in this document, a_i(t)
may contain ineligible arrivals that need firstly to be converted (or
sorted) into eligible arrivals, e.g., by method of regulation
(Section 5) or latency compensation (Section 6), and then processed
by the EDF scheduler.
3.2.1. Conditions for Leaky Bucket Constraint Function
Assume that we want to support n delay levels (d_1, d_2,..., d_n) in
the network, and the traffic arrival constraint function of each
delay level d_i is the leaky bucket arrival curve A_i(t) = b_i + r_i
* t. Equation-1 can be expressed as:
b_1 <= C*d_1 - M
b_1 + b_2 + r_1*(d_2-d_1) <= C*d_2 - M
b_1 + b_2 + b_3 + r_1*(d_3-d_1) + r_2*(d_3-d_2) <= C*d_3 - M
... ...
sum(b_1+...+b_n) + r_1*(d_n-d_1) + r_2*(d_n-d_2) + ... +
r_n_1*(d_n-d_n_1) <= C*d_n - M
where, C is the service rate of the EDF scheduler, M is the maximum
size of the interference packet.
Note that the preset value of b_i does not depend on r_i, but r_i
generally refers to b_i (and burst interval) for setting. For
example, the preset value of r_i may be small, while the value of b_i
may be large. Such parameter design is more suitable for
transmitting traffic with large service burst interval, large service
burst size, but small bandwidth requirements.
Peng, et al. Expires 9 February 2025 [Page 9]
Internet-Draft Deadline Queueing Mechanism August 2024
An extreme example is that the preset r_i of each level d_i is close
to 0 (this is because the burst interval of the served flow is too
large, e.g., one hour or one day), but the preset b_i is close to the
maximum value (e.g., b_1 = C*d_1 - M), then when the concurrent flow
of all delay levels is scheduled, the time 0~d_1 is all used to send
the burst b_1, the time d_1~d_2 is all used to send the burst b_2,
the time d_2~d_3 is all used to send the burst b_3, and so on.
However, the typical allocation scheme is that the preset r_i of each
level d_i will divide C roughly equally. For example, we may firstly
pre-allocate b_1 = C*d_1 - M, r_1 = C/n; Then recursively pre-
allocate b_2 = C*(d_2-d_1)*(n-1)/n, r_2 = C/n; And so on. The pre-
allocated parameters b_i and r_i of each level d_i constitute the
delay resources of that level of the link. A path can reserve
required burst and bandwidth from delay resources of the specific
delay level d_i, and the reservation is successful only if the two
resources are successfully reserved at the same time. As long as
neither b_i nor r_i is free, the delay resource of level d_i is
exhausted.
Alternatively, a more tight allocation scheme is to not preset the
parameters of A_i(t), but to dynamically accumulate the parameters of
A_i(t) based on the actual flows setup demand, and always check
whether the schedulability condition is met based on the updated
A_i(t) during the flow setup procedure. In this case, it is still
necessary to set a resource limit for each delay level to prevent the
flows of a certain delay level from consuming all resources. For
example, we may set the resource limit of each delay level d_i to
b_i_limit = C*(d_i - d_(i-1)) - M, r_i_limit = C/n. In this case,
the dynamically updated b_i and r_i should be treated as utilized
resources, and participate in schedulability condition checks.
Note that for some delay level d_i, its resource may be explicitly
set to empty, i.e., b_i = 0, r_i = 0. This brings flexibility, and
resources can be freed up for later delay levels with lower priority
to use.
In a specific scenario, if the ideal arrival packet interval (by the
method of re-shaping or latency compensation) of all service flows is
large, and the maximum delay level d_n chosen is not larger than any
packet interval of any service flow, the schedulability condition can
be further simplified as follows:
b_1 <= C*d_1 - M, r_1 = b_1 / d_n;
b_1 + b_2 <= C*d_2 - M, , r_2 = b_2 / d_n;
b_1 + b_2 + b_3 <= C*d_3 - M, r_3 = b_3 / d_n;
Peng, et al. Expires 9 February 2025 [Page 10]
Internet-Draft Deadline Queueing Mechanism August 2024
... ...
sum(b_1+...+b_n) <= C*d_n - M, r_n = b_n / d_n;
The above simplified condition implies that the total number of
bursts contained within any time interval d_n does not exceed
sum(b_1+...+b_n). This is true because for any flow i it never
contains two packets in a single time interval d_n. In this case, it
can support a larger service scale than the original condition.
3.2.2. Schedulability Condition Analysis for On-time Mode
Compared with in-time mode, on-time mode is non-work-conserving,
which can be considered as the combination of damper and EDF
scheduler. On-time scheduling mode applied on a flow try to maintain
the time interval between any adjacent packets of that flow to be
consistent with the regulated interval on the flow entrance node.
The maintenance of time intervals does not lead to an increase in the
bandwidth occupied by that flow and cause the arrival curve to
violate the traffic constraint function. So that the schedulability
condition (i.e., Equation-1) can also be applied to on-time
scheduling mode. See Section 11 for more information about jitter
control.
3.3. Buffer Size Design
The service rate of the EDF scheduler, termed as C, can reach the
link rate, but generally only needs to be configured as part of the
link bandwidth, such as 50%. Allow hierarchical scheduling, for
example, the EDF scheduler may participate in higher-level WFQ
scheduling along with other schedulers.
If flows are rate-controlled (i.e., reshaping is done inside the
network, or on-time mode is applied), the maximum depth of PIFO
should be C * d_n, where d_n is the maximum delay level. Otherwise,
more buffer is necessary to store the accumulated bursts, that is,
the PIFO zone where the distance from HoQ exceeds the maximum delay
level is just used to store accumulated bursts. Please refer to
Section 16 for more considerations.
4. Rotation Priority Queues
[RPQ] described rotating priority queues, and the priority
granularity of the queue is the same as that of the flows. If the
deadline of the flow is used as priority, it requires a lot of
priority and corresponding queues, with scalability issues.
Therefore, this section provides rotating priority queues with count-
down time range whose rotation interval is more refined, with the
Peng, et al. Expires 9 February 2025 [Page 11]
Internet-Draft Deadline Queueing Mechanism August 2024
following characteristics:
* Each queue has CT (Count-down Time) that is decreased by RTI
(Rotation Time Interval). The CT difference between two adjacent
queues is CTI (CT Interval). RTI must be less than or equal to
CTI, with CTI = K * RTI, where the natural number K >= 1.
* The smaller the CT, the higher the priority. At the beginning,
all queues have different initial CT values, i.e., staggered from
each other, e.g., one queue has the minimum CT value (termed as
MIN_CT), and one queue has the maximum CT value (termed as
MAX_CT), and the CT values of all queues increase equally by
CTI.It should be noted that CT is just the countdown of the HoQ,
and the countdown of the end of the queue (EoQ) is near CT+CTI.
So the CT attribute of a queue is actually a range [CT, CT+CTI).
* For a queue whose CT is MIN_CT, after a new round of CTI, its CT
will become MIN_CT - CTI and immediately return to MAX_CT.
The above CTI, RTI, MIN_CT and MAX_CT value should be choosed
according to the hardware capacity. Each link can independently use
different CTI. The general principle is that the larger bandwidth,
the smaller CTI. The CTI must be designed large enough to include
interference delay caused by a single low priority packet with
maximum size.
The choose of RTI should consider the latency granularity of various
DetNet flows, so that CT updated per RTI can match the delay
requirements of different flows. One implementation may not choose
to let CT be actually updated at the granularity of RTI, but at the
granularity of CTI. For example, the elapsed time within CTI can be
recorded, and (cur_CT - elapsed_time) can be used as the actual CT of
the queue, where cur_CT is the current CT of the queue that has not
been updated yet. Although the update of cur_CT is slow, the actual
CT is sensitive enough.
According to different scheduling mode configured to the RPQ, MIN_CT
may be designed to different values. For in-time mode, MIN_CT may be
0. For on-time mode with option E|D decoupling (see Section 11),
MIN_CT may also be 0, assuming that the packet allows departure from
pre-scheduler, the time it takes to send to post-scheduler can be
ignored. For on-time mode with option E+D integration, MIN_CT may be
-N*CTI, where N is the amount of delay levels, considering the
transmission time to the output link cannot be ignored.
A specific example of RPQ configured with in-time scheduling mode is
depicted in Figure 1.
Peng, et al. Expires 9 February 2025 [Page 12]
Internet-Draft Deadline Queueing Mechanism August 2024
+------------------------------+ +------------------------------+
| RPQ Group: | | RPQ Group: |
| queue-1(CT=50us) ###### | | queue-1(CT=49us) ###### |
| queue-2(CT=40us) ###### | | queue-2(CT=39us) ###### |
| queue-3(CT=30us) ###### | | queue-3(CT=29us) ###### |
| queue-4(CT=20us) ###### | | queue-4(CT=19us) ###### |
| queue-5(CT=10us) ###### | | queue-5(CT= 9us) ###### |
| queue-6(CT=0us) ###### | | queue-6(CT=-1us) ###### |
+------------------------------+ +------------------------------+
+------------------------------+ +------------------------------+
| Other Queue Group: | | Other Queue Group: |
| queue-7 ############ | | queue-7 ############ |
| queue-8 ############ | | queue-8 ############ |
| queue-9 ############ | | queue-9 ############ |
| ... ... | | ... ... |
+------------------------------+ +------------------------------+
-o----------------------------------o------------------------------->
T0 T0+1us time
Figure 1: Example of RPQ Groups
In this example, the CTI for RPQ group is configured to 10us.
Queue-1 ~ queue-6 are members of RPQ group. Each queue has its
initial CT attribute, and the CT of all queues are staggered from
each other. For example, the CT of queue-1 is 50us (MAX_CT), the CT
of queue-2 is 40uS, ..., the CT of queue-6 is 0 (MIN_CT).
Suppose the scheduling engine initiates a rotation timer with a time
interval of 1us, i.e., CTI = 10 * RTI in this case. As shown in the
figure, at T0 + 1us, the CT of queue-1 becomes 49us, the CT of
queue-2 becomes 39us, etc.
At T0 + 10us, the CT of queue-6 will return to 50us (MAX_CT).
Note that the minimum D requested by a DetNet flow should not be
smaller than d_1+F, where d_1 is the most urgent delay level, F is
the intra node forwarding delay. Therefore any packets with in-time
scheduling should have Q (i.e., D + E - F) that is not be smaller
than d_1, and should never be inserted to a queue whose CT is
negative.
Peng, et al. Expires 9 February 2025 [Page 13]
Internet-Draft Deadline Queueing Mechanism August 2024
4.1. Alternate Queue Allocation Rules
There may be extreme scenarios that multiple delay levels of eligible
bursts arrive sequentially, with lower priority burst arriving first
and higher priority burst arriving later, and then simultaneously
releasing flood. In this case, it is necessary to ensure that the
higher priority burst is sent first to meet its deadline.
Therefore it may further let a RPQ queue (act as the virutal parent
queue) contain multiple sub-queues, each for a delay level. The
physical sub-queue with small delay level (e.g., 10us) is ranked
before the physical sub-queue with large delay level (e.g., 20us).
Packets are actually stored in the physical sub-queues. That is,
packets belonging to different delay levels are inserted into
different sub-queues and protected. In this way, for two packets
with the same Q but different D, we can decide to firstly schedule
the packet with the smallest D.
This alternate queue allocation rule enables eligible arrivals always
have a place to store, avoiding conflicts in local positions of the
RPQ queue group and causing overflow.
4.2. Scheduling Mode for RPQ
A RPQ group may be configured as either in-time or on-time scheduling
mode, but cannot support both modes simultaneously.
In the in-time scheduling mode, in all non empty queues, the packets
in each queue are sequentially sent in the order of high priority
queue to low priority queue. The actual bandwidth consumed by the
scheduler may exceed its set service rate C.
In the on-time scheduling mode, only in all non empty queues with CT
<= 0, the packets in each queue are sequentially sent in the order of
high priority queue to low priority queue.
For a virtual parent queue that is allowed to be sent, for the
multiple non empty physical sub-queues it contains, packets are
sequentially sent from the non empty physical sub-queues along the
direction from the physical sub-queues with small delay levels to the
physical sub-queues with large delay levels. Only when a physical
sub-queue is cleared can the next non empty physical sub-queue be
sent.
4.3. Schedulability Condition for RPQ
In this section, we discuss the schedulability condition based on RPQ
with in-time scheduling mode firstly.
Peng, et al. Expires 9 February 2025 [Page 14]
Internet-Draft Deadline Queueing Mechanism August 2024
Suppose for any delay level d_i, the corresponding accumulated
constraint function is A_i(t), and let d_i < d_(i+1). Suppose for
any planned residence time D_i, the the corresponding constraint
function is A'_i(t). For simplicity, we take intra node forwarding
delay F as 0. Then the schedulability condition is:
* A_1(t-d_1) + sum{A_i(t+CTI-d_i) for all i>=2} <= C*t, if a d_i
contains only one D_i. (Equation-2)
* sum{A_i(t+CTI-d_i) for all i>=1} <= C*t, if d_i contains multiple
D_i. (Equation-3)
where CTI is the CT interval between adjacency queue, C is service
rate of the EDF scheduler.
The proof is similar with that in [RPQ], except that the rotation
step is fine-grained by RTI and the priority of each queue is CT
range. Please refer to Appendix A for the proof.
Note that the key difference between the above two conditions (i.e.,
Equation-2, Equation-3) and one based on sorted queue (i.e.,
Equation-1) is the CTI factor.
Other common considerations are the same as Section 3.2.
4.3.1. Schedulability Condition for Alternate QAR
According to Section 4.1, a RPQ queue may further contain multiple
sub-queues, each for a delay level. Under the same parent queue, all
sub-queues are sorted in descending order of delay level. In this
case, the precise workload should exclude packets with higher delay
levels than the observed packet.
In the case that d_i contains only one D_i, the schedulability
condition is Equation-1.
In the case that d_i contains multiple D_i, the schedulability
condition is still Equation-3.
Please refer to Appendix B for the proof.
4.3.2. Conditions for Leaky Bucket Constraint Function
Assume that we want to support delay levels (d_1, d_2,..., d_n) in
the network, and the traffic arrival constraint function of each
delay level d_i is the leaky bucket arrival curve A_i(t) = b_i + r_i
* t. Equation-2 can be expressed as:
Peng, et al. Expires 9 February 2025 [Page 15]
Internet-Draft Deadline Queueing Mechanism August 2024
b_1 <= C*d_1 - M
b_1 + b_2 + (r_1+r_2)*CTI <= C*d_2 - M
b_1 + b_2 + b_3 + (r_1+r_2)*2*CTI + r_3*CTI <= C*d_3 - M
... ...
sum(b_1+...+b_n) + (r_1+r_2)*(n-1)*CTI + r_3*(n-2)*CTI + ... +
r_n*CTI <= C*d_n - M
where, C is the service rate of the EDF scheduler, M is the maximum
size of the interference packet.
Equation-3 can be expressed as:
b_1 + r_1*CTI <= C*d_1 - M
b_1 + b_2 + r_1*2*CTI + r_2*CTI <= C*d_2 - M
b_1 + b_2 + b_3 + r_1*3*CTI + r_2*2*CTI + r_3*CTI <= C*d_3 - M
... ...
sum(b_1+...+b_n) + r_1*n*CTI + r_2*(n-1)*CTI + ... + r_n*CTI <=
C*d_n - M
Similarly, in a specific scenario, if the ideal arrival packet
interval (by the method of re-shaping or latency compensation) of all
service flows is large, and the maximum delay level d_n chosen is not
larger than any packet interval of any service flow, the above two
schedulability conditions can be further simplified as follows:
b_1 <= C*d_1 - M, r_1 = b_1 / d_n;
b_1 + b_2 <= C*d_2 - M, , r_2 = b_2 / d_n;
b_1 + b_2 + b_3 <= C*d_3 - M, r_3 = b_3 / d_n;
... ...
sum(b_1+...+b_n) <= C*d_n - M, r_n = b_n / d_n;
Peng, et al. Expires 9 February 2025 [Page 16]
Internet-Draft Deadline Queueing Mechanism August 2024
4.3.3. Schedulability Condition Analysis for On-time Mode
Compared with in-time mode, on-time mode is non-work-conserving,
which can be considered as the combination of damper and EDF
scheduler. On-time scheduling mode applied on a flow try to maintain
the time interval between any adjacent packets of that flow to be
consistent with the regulated interval on the flow entrance node.
The maintenance of time intervals does not lead to an increase in the
bandwidth occupied by that flow and cause the arrival curve to
violate the traffic constraint function. So that the schedulability
condition (i.e., Equation-2/3) can also be applied to on-time
scheduling mode. See Section 11 for more information about jitter
control.
4.4. Buffer Size Design
An implementation may let all queues share the common buffer.
Especially if alternet QAR (Section 4.1) is applied, the actual
buffer cost of a virtual parent queue is contributed by all the
physical sub-queues it contains. The actual buffer cost of each
physical sub queue is dynamically allocated based on whether there is
a packet inserted. According to Section 4.3, the maximum buffer cost
of a physical sub-queue may reach the upper limit of burst resources
for the corresponding delay level.
If flows are rate-controlled (i.e., reshaping is done inside the
network, or on-time scheduling mode is applied), the MAX_CT may be
designed as the maximum delay level, and total necessary buffer
shared by all queues should be C * d_n, where C is the service rate
and d_n is the maximum delay level. Otherwise, MAX_CT should be
larger than the maximum delay level, and with more necessary buffer,
to store the accumulated bursts, that is, all the queues with CT
larger than the maximum delay level are just used to store
accumulated bursts. Please refer to Section 16 for more
considerations.
5. Reshaping
Reshaping per flow inside the network, as described in [RFC2212], is
done at all heterogeneous source branch points and at all source
merge points, to restore (possibly distorted) traffic's shape to
conform to the TSpec. Reshaping entails delaying packets until they
are within conformance of the TSpec.
Peng, et al. Expires 9 February 2025 [Page 17]
Internet-Draft Deadline Queueing Mechanism August 2024
A network element MUST provide the necessary buffers to ensure that
conforming traffic is not lost at the reshaper. Note that while the
large buffer makes it appear that reshapers add considerable delay,
this is not the case. Given a valid TSpec that accurately describes
the traffic, reshaping will cause little extra actual delay at the
reshaping point (and will not affect the delay bound at all).
Maintaining a dedicated shaping queue per flow can avoid burstiness
cascading between different flows with the same traffic class, but
this approach goes against the design goal of packet multiplexing
networks. [IR-Theory] describes a more concise approach by
maintaining a small number of interleaved regulators (per traffic
class and incoming port), but still maintaining the state of each
flow. With this regulator, packets of multiple flows are processed
in one FIFO queue and only the packet at the head of the queue is
examined against the regulation constraints of its flow. However, as
the number of flows increases, the IR operation may become burdensome
as much as the per-flow reshaping.
For any observed EDF scheduler in the network, when the traffic
arriving from all incoming ports is always reshaped, then these flows
comply with their arrival constraint functions, which is crucial for
the schedulability conditions of EDF scheduling. Based on this, it
can quantify the delay resource pool which is open and reserved for
DetNet flows.
6. Latency Compensation
[RFC9320] presents a latency model for DetNet nodes. There are six
type of delays that a packet can experience from hop to hop. The
processing delay (type-4), the regulator delay (type-5) , the
queueing subsystem delay (type-6), and the output delay (type-1)
together contribute to the residence time in the node.
In this document, the residence time in the node is simplified into
two parts: the first part is to lookup the forwarding table when the
packet is received from the incoming port (or generated by the
control plane) and deliver the packet to the line card where the
outgoing port is located; the second part is to store the packet in
the queue of the outgoing port for transmission. These two parts
contribute to the actual residence time of the packet in the node.
The former can be called forwarding delay (termed as F) and the
latter can be called queueing delay (termed as Q). The forwarding
delay is related to the chip implementation and is generally constant
(with a maximum value); The queueing delay is unstable.
Peng, et al. Expires 9 February 2025 [Page 18]
Internet-Draft Deadline Queueing Mechanism August 2024
6.1. Get Accumulated Residence Time Deviation
The accumulated residence time deviation, also termed as latency
deviation (E), equals accumulated planned residence time minus
accumulated actual residence time. This value can be zero, positive,
or negative.
The accumulated planned residence time of the packet refers to the
sum of the planned residence time of all upstream nodes before the
packet is transmitted to the current node. The accumulated actual
residence time of the packet, refers to the sum of the actual
residence time of all upstream nodes before the packet is transmitted
to the current node.
In the case of in-time scheduling, E may be a very large positive
value. While in the case of on-time scheduling, E may be 0, or a
small value close to 0.
The setting of the latency deviation (E) of the packet needs to be
friendly to the chip for reading and writing. For example, it should
be designed as a fixed position in the packet. The chip may support
flexible configuration for that position.
[I-D.peng-6man-delay-options] defined the method for carrying the
latency deviation (E) in the IPv6 Hop-by-Hop Options Header.
[I-D.pb-6man-deterministic-crh], [I-D.p-6man-deterministic-eh]
defined methods for carrying the latency deviation (E) in the IPv6
Routing Header.
6.2. Get Allowable Queueing Delay
When an EDF scheduler receives a packet, it can calculate allowable
queueing delay (Q) for the packet. Specifically, it can first get
the latency deviation (E), and add it to the planned residence time
(D) of the packet at this node to obtain the adjustment residence
time, and then deduct the actual forwarding delay (F) of the packet
in the node, to obtain the allowable queueing delay (Q) for that
packet.
* Q = D + E - F
The scheduler selects a buffer position (e.g., queue-id, or rank) for
the packet based on Q.
Note that one implementation may calculate Q at incoming port and
determine the buffer position of the outgoing port. In this case, Q
= D + E, and a buffer position indication may be notified from the
incoming port to the outgoing port.
Peng, et al. Expires 9 February 2025 [Page 19]
Internet-Draft Deadline Queueing Mechanism August 2024
In detailed, assume that the current node in a deterministic path is
h, all upstream nodes are from 1 to h-1. For any node i, denote the
planned residence time as D[i], the actual residence time as R[i],
the input latency deviation (contributed by all upstream nodes) as
E[i], the forwarding delay intra-node as F[i], then the allowable
queueing delay (Q) of the packet on node h is:
Q[h] = D[h] + E[h] - F[h]
E[h] = D[h-1] + E[h-1] - R[h-1]
D[0], E[0], R[0] = 0
6.3. Scheduled by Allowable Queueing Delay
The packet will be sheduled based on its Q, that is, the packet is
scheduled based on latency compensation contributed by E, instead of
only D. The earliest literature similar to the idea of latency
compensation based on E can be found in [Jitter-EDF].
The core stateless latency compensation can achieve the effect of
reshaping per flow to get the eligible arrivals pattern. Q can be
used to sort ineligible arrvials of one delay level and prevent them
from interferring with the scheduling of eligible arrvials of other
delay levels.
Firstly, at the flow (e.g., flow i) entrance node, all packets (after
regulation) of flow i will be released to the EDF scheduler one after
another at different time (termed as ideal arrival time), but with
the same allowable queueing delay (Q), with initial E = 0, i.e., Q =
D, assuming no link propagation delay and intra-node forwarding delay
for simplicity. We denote this arrival pattern faced by the
shceduler on the flow entrance node as arrival_pattern_0, which
contains a sequence of packets with varaint of intervals between
adjacent packets. We say that arrival_pattern_0 is eligible arrivals
because its arrival curve is less than the constraint function
A_i(t). For any packet p in arrival_pattern_0, assuming its ideal
arrival time is t_p_0.
Then, we can get arrival_pattern_1 = arrival_pattern_0 + D that is
also eligible arrivals, where, arrival_pattern_0 + D means that the
ideal arrival time (at the scheduler of flow entrance node) of each
packet in arrival_pattern_0 is added with D. In fact,
arrival_pattern_1 is the eligible arrivals on the second node. That
is, the second node may recover the eligible arrivals
arrival_pattern_1 from the actual arrivals with the help of latency
compensation, and then to schedule based on arrival_pattern_1. How
did arrival_pattern_1 recover? For any packet p, assuming it
Peng, et al. Expires 9 February 2025 [Page 20]
Internet-Draft Deadline Queueing Mechanism August 2024
experiences an actual queuing delay q on the flow enrance node, and
will actually arrive at the second node at time t_p_0 + q, with E = D
- q carried in the sending packet. The second node will recover the
eligible arrival time of packet p by, eligible arrival time = actual
arrival time + E = t_p_0 + q + D - q = t_p_0 + D. Therefore
arrival_pattern_1 is recovered.
Similarly, the third node may recover arrival_pattern_2 =
arrival_pattern_0 + 2*D, and the fourth node may recover
arrival_pattern_3 = arrival_pattern_0 + 3*D, and so on. On any node
h, packet p will be sorted in the scheduler queue based on its
eligible arrival time plus D, i.e., ideal departure time, for
scheduling.
Because the scheduler always schedules based on eligible arrivals,
its scheduling power will not be overwhelmed by actual arrivals that
may include burst accumulation.
We may think the packets sorted in the queue with ideal departure
time as a virtual regulation, because the rank distance between the
adjacent packets of the flow i is exactly maintained consistently
with the corresponding regulated interval between these two adjacent
packets on the flow entrance node. There is no need to require that
this virtual regulation and a real regulation component must have
exactly the same pattern, as long as each pattern is less than the
arrival constraint function A_i(t).
Although, lantency compensation has the effect of reshaping, but it
is not equivalent to reshaping. Considering an accumulated bursts
that violates the traffic constraint function and arrives at a node,
if reshaping is used, it will substantially introduce shaping delay
for the ineligible bursts, which will then enter the queueing
subsystem. While if latency compensation is used, this ineligible
bursts will only be penalized with a larger Q and tolerated to be
placed in the queueing sub-system, and in the case of in-time mode it
may be immediately sent if higher priority queues are empty.
Note that the premise of latency compensation is that a flow must be
based on a fixed explicit path. If multiple packets from the same
flow arrive at the intermediate node via multiple paths with
different propagation lengths, even if these packets are all
eligible, bursts accumulation may still form and cannot even be
punished.
Peng, et al. Expires 9 February 2025 [Page 21]
Internet-Draft Deadline Queueing Mechanism August 2024
7. Option-1: Reshaping plus Sorted Queue
As shown in Figure 2, a receivd packet is inserted to the PIFO queue
according to rank = A + D - F, where, A is the time that packet
arrived at the scheduler, i.e., arrive_time_S in the figure.
Depending on the situation of the accumulated burst arrived at the
input port, different packets may face different shaping delays. The
shaper will convert the input ineligible arrivals pattern (if
possible) into an eligible arrivals pattern. Here, D - F may be
denoted as the allowable queueing delay Q.
+---------------------------------------------------+
| +---+ +-+ +--------+ +-----------+ |
| | | |X| | | | Scheduler | |
Input | | S | |X| | | | (PIFO) | | Output
port -O -> | & | -> |X| -> | Shaper | -> | top->[==] | -> O- port
| | F | |X| | | | [==] | |
| | | |X| | | |rank->[==] | |
| | | |X| | | | [==] | |
| +---+ +-+ +--------+ +-----------+ |
+---------------------------------------------------+
| |
------o----------------------------------o------------------->
arrive_time_I arrive_time_S time
|<-------- F ------->|<---- S ---->|<----- Q ------>|
Figure 2: Reshaping plus Sorted Queue
Enqueue rule:
* For two packets with different rank, the packet with a smaller
rank is closer to the head of the queue.
* For two packets with the same rank, the packet with a smaller D is
closer to the head of the queue.
* For two packets with the same rank and D, the packet that arrive
at the scheduler first is closer to the head of the queue.
The planned residence time (D) should be carried in the packet.
The scheduling mode (in-time or on-time) should also be carried in
the packet, and used to insert packet into PIFO with the
corresponding scheduling mode.
Dequeue rule:
Peng, et al. Expires 9 February 2025 [Page 22]
Internet-Draft Deadline Queueing Mechanism August 2024
* As mentioned in Section 3.1, for a PIFO with in-time scheduling
mode, as long as the queue is not empty, packets are always
departured from the HoQ for transmission; while for PIFO with on-
time scheduling mode, only if the queue is not empty and the rank
of the HoQ packet is equal to or earlier than the current system
time, the HoQ packet can be sent.
However, in this option the dequeue rule of on-time mode can not
guarantee jitter, due to lack of factor E to absorb jitter per hop.
The dequeue rule of on-time mode only controls the starting time when
packets are allowed to be scheduled, but cannot guarantee that
different packets have the same queuing delay.
8. Option-2: Reshaping plus RPQ
As shown in Figure 3, a receivd packet is inserted to the appropriate
RPQ queue with specific CT to meet CT <= Q < CT+CTI when the packet
arrived at the scheduler, where Q = D - F. Depending on the
situation of the accumulated burst arrived at the input port,
different packets may face different shaping delays. The shaper will
convert the input ineligible arrivals pattern (if possible) into an
eligible arrivals pattern.
+----------------------------------------------------+
| +---+ +-+ +--------+ +------------+ |
| | | |X| | | | Scheduler | |
Input | | S | |X| | | | (RPQ) | | Output
port -O -> | & | -> |X| -> | Shaper | -> | CT1 #### | -> O- port
| | F | |X| | | | CT2 #### | |
| | | |X| | | |Q->CT3 #### | |
| | | |X| | | | CT4 #### | |
| +---+ +-+ +--------+ +------------+ |
+----------------------------------------------------+
| |
------o----------------------------------o-------------------->
arrive_time_I arrive_time_S time
|<-------- F ------->|<---- S ---->|<------ Q ------>|
Figure 3: Reshaping plus RPQ
Enqueue rule:
* For a packet with Q, select the target RPQ queue (i.e., the
virtual parent queue) with corresponding CT, that meet CT <= Q <
CT+CTI.
Peng, et al. Expires 9 February 2025 [Page 23]
Internet-Draft Deadline Queueing Mechanism August 2024
* Under the selected virtual parent queue, select the target
physical sub-queue with corresponding delay level d_i, which is
closest to D-F and not greater than D-F.
The planned residence time (D) should be carried in the packet.
The scheduling mode (in-time or on-time) should also be carried in
the packet, and used to insert packet into RPQ with the corresponding
scheduling mode.
Dequeue rule:
* As mentioned in Section 4.2, for a RPQ group with in-time
scheduling mode, in all non empty queues, the packets in each
queue are sequentially sent in the order of high priority queue to
low priority queue; while for a RPQ group with on-time scheduling
mode, only in all non empty queues with CT <= 0, the packets in
each queue are sequentially sent in the order of high priority
queue to low priority queue.
However, in this option the dequeue rule of on-time mode can not
guarantee jitter, due to lack of factor E to absorb jitter per hop.
The dequeue rule of on-time mode only controls the starting time when
packets are allowed to be scheduled, but cannot guarantee that
different packets have the same queuing delay.
9. Option-3: Latency Compensation plus Sorted Queue
As shown in Figure 4, a receivd packet is inserted to the PIFO queue
according to rank = A1 + E + D, or rank = A2 + E + D - F, where, A1
is the time that packet arrived at the input port (i.e.,
arrive_time_I in the figure), A2 is the time that packet arrived at
the scheduler (i.e., arrive_time_S in the figure). Note that E is
initially 0 on the flow entrance node, and generally not 0 on other
nodes and will update per hop. Depending on the situation of the
accumulated burst arrived at the input port, different packets may
have different input latency deviation E. Latency compensation will
convert the input ineligible arrivals pattern (if possible) into an
eligible arrivals pattern. Here, E + D - F may be denoted as the
allowable queueing delay Q.
Peng, et al. Expires 9 February 2025 [Page 24]
Internet-Draft Deadline Queueing Mechanism August 2024
+-----------------------------------------+
| +---+ +-+ +-----------+ |
| | | |X| | Scheduler | |
Input | | S | |X| | (PIFO) | | Output
port -O --> | & | --> |X| --> | top->[==] | --> O- port
| | F | |X| | [==] | |
| | | |X| |rank->[==] | |
| | | |X| | [==] | |
| +---+ +-+ +-----------+ |
+-----------------------------------------+
| |
------o-----------------------o------------------->
arrive_time_I arrive_time_S time
|<--------- F --------->|<------ Q ------>|
Figure 4: Latency Compensation plus Sorted Queue
The planned residence time (D) and latency deviation (E) should be
carried in the packet.
The enqueue and dequeue operations are the same as Section 7.
In this option the dequeue rule of on-time mode can guarantee jitter
with the help of factor E to absorb jitter per hop. See Section 11
for more information.
9.1. Packet Disorder Considerations
Suppose that two packets, P1, P2, are generated instantaneously from
a specific flow at the source, and the two packets have the same
planned residence time. P1 may face less interference delay than P2
in their journey. When they arrive at an intermediate node in turn,
P2 will have less allowable queueing delay (Q) than P1 to try to stay
close to P1 again. It should be noted that to compary who is ealier
is based on the time arriving at the scheduler plus packet's Q. The
time difference between the arrival of two packets at the scheduler
may not be consistent with the difference between their Q. It is
possible to get an unexpected comparision result.
As shown in Figure 5, P1 and P2 are two back-to-back packets
belonging to the same flow. The arrival time when they are received
on the scheduler is shown in the figure. Suppose that the Q values
of two adjacent packets P1 and P2 are 40us and 39us, and arrive at
the scheduler at time T1 and T2 respectively. P1 will be sorted
based on T1 + 40us, while P2 will be sorted based on T2 + 39us.
Ideally, T2 should be T1 + 1us. However, this may be not the case.
For example, it is possible that T2 = T1 + 0.9us, Q1 = 40, Q2 = 39.1,
but just because the calculation accuracy of Q1 and Q2 is
Peng, et al. Expires 9 February 2025 [Page 25]
Internet-Draft Deadline Queueing Mechanism August 2024
microseconds, so they are, e.g., with half-adjust, approximately 40
us and 39 us, respectively. This means that P2 will be sorted before
P1 in the PIFO, resulting in disorder.
packets arrived later
packets arrived earlier |
| |
V V
--------+-----------------------------------------------+---------
... ... | .P1.................P2....................... | ... ...
--------+-----------------------------------------------+---------
P1.Q=40us
P2.Q=39us
| |
--------o---------------------o--------------------------------->
T1 T2 (=T1+0.9us) time
| ___________________|
| |
v v
PIFO ##############################################################
top
Figure 5: Disorder Illustration of PIFO
DetNet architecture [RFC8655] provides Packet Ordering Function
(POF), that can be used to solve the above disorder problem caused by
the latency compensation.
Alternatively, Section 11 provides E|D decoupling method to firstly
absorb latency deviation E by the pre-scheduler which may maintain
FIFO queue per incoming port plus delay level. In this case, packets
from the same flow will only determine the damping delay, but not the
position, in the FIFO based on latency deviation E, to avoid
disorder. Latency deviation E no longer works in the post-scheduler.
Note that in practical situations, two back-to-back packets of the
same flow are generally evenly distributed within the burst interval
by regulation, which means that the distance between these two
packets is generally much greater than the calculation accuracy
mentioned above, meaning that the disordered phenomenon will not
really occur. For example, the regulated result meets a Length Rate
Quotient (LRQ) constraint, and the time interval between two
consecutive packets of size l_i and l_j should be at least l_i/r,
where r is the flow rate (i.e., the reserved bandwidth of the flow).
This can be done by LRQ based regulation, or enhanced leaky bucket
based regulation, depending on implementation.
Peng, et al. Expires 9 February 2025 [Page 26]
Internet-Draft Deadline Queueing Mechanism August 2024
10. Option-4: Latency Compensation plus RPQ
As shown in Figure 6, a receivd packet is inserted to the appropriate
RPQ queue with specific CT to meet CT <= Q < CT+CTI when the packet
arrived at the scheduler, where Q = D + E - F. Depending on the
situation of the accumulated burst arrived at the input port,
different packets may have different input latency deviation E.
Latency compensation will convert the input ineligible arrivals
pattern (if possible) into an eligible arrivals pattern.
+-------------------------------------------+
| +---+ +-+ +-------------+ |
| | | |X| | Scheduler | |
Input | | S | |X| | (RPQ) | | Output
port -O --> | & | --> |X| --> | CT1 #### | --> O- port
| | F | |X| | CT2 #### | |
| | | |X| |Q-> CT3 #### | |
| | | |X| | CT4 #### | |
| +---+ +-+ +-------------+ |
+-------------------------------------------+
| |
------o-----------------------o---------------------->
arrive_time_I arrive_time_S time
|<--------- F --------->|<------ Q ------>|
Figure 6: Latency Compensation plus RPQ
The planned residence time (D) and latency deviation (E) should be
carried in the packet.
The enqueue and dequeue operations are the same as Section 8.
In this option the dequeue rule of on-time mode can guarantee jitter
with the help of factor E to absorb jitter per hop. See Section 11
for more information.
Figure 7 depicts an example of packets inserted to the RPQ queues.
Peng, et al. Expires 9 February 2025 [Page 27]
Internet-Draft Deadline Queueing Mechanism August 2024
P2 P1 +------------------------------+
+--------+ +--------+ | RPQ Group: |
| D=20us | | D=30us | | queue-1(CT=45us) ###### |
| E=15us | | E=-8us | +--+ | queue-2(CT=35us) ###### |
+--------+ +--------+ |\/| | queue-3(CT=25us) ###### |
------incoming port-1------> |/\| | queue-4(CT=15us) ###### |
|\/| | queue-5(CT=5us) ###### |
P4 P3 |/\| | queue-6(CT=-5us) ###### |
+--------+ +--------+ |\/| | queue-7(CT=-15us)###### |
| | | D=30us | |/\| | ... ... |
+--------+ | E=-30us| |\/| +------------------------------+
+--------+ |/\|
------incoming port-2------> |\/| +------------------------------+
|/\| | Other Queue Group: |
P6 P5 |\/| | queue-8 ############ |
+--------+ +--------+ |/\| | queue-9 ############ |
| | | D=40us | |\/| | queue-10 ############ |
+--------+ | E=40us | |/\| | ... ... |
+--------+ +--+ +------------------------------+
------incoming port-3------> ----------outgoing port--------->
-o----------------------------------o------------------------------>
arrival time +F time
Figure 7: Time Sensitive Packets Inserted to RPQ
As shown in Figure 7, the node successively receives six packets from
three incoming ports, among which packet 1, 2, 3 and 5 have
corresponding deadline information, while packet 4 and 6 are best-
effort packets. These packets need to be forwarded to the same
outgoing port. It is assumed that they arrive at the line card where
the outgoing port is located at almost the same time after the
forwarding delay (F = 5us). At this time, the queue status of the
outgoing port is shown in the figure. Then:
* The allowable queueing delay (Q) of packet 1 is 30 - 8 - 5 = 17us,
and it will be put into queue-4 (its CT is 15us), meeting the
condition that Q is in the range [15, 25).
* The allowable queueing delay (Q) of packet 2 is 20 + 15 - 5 =
30us, and it will be put into queue-3 (its CT is 25us), meeting
the condition that Q is in the range [25, 35).
* The allowable queueing delay (Q) of packet 3 is 30 - 30 - 5 =
-5us, and it will be put into queue-6 (its CT is -5us), meeting
the condition that Q is in the range [-5, 5).
Peng, et al. Expires 9 February 2025 [Page 28]
Internet-Draft Deadline Queueing Mechanism August 2024
* The allowable queueing delay (Q) of packet 5 in the node is 40 +
40 - 5 = 75us, and the queue it is placed on is not shown in the
figure (such as a hierarchical queue).
* Packets 4 and 6 will be put into the non-deadline queue in the
traditional way.
According to Section 4.3, An eligible packet (i.e., E = 0) from a
specific delay level, even at the end of the inserted queue, can
ensure that it does not exceed its deadline, which is the key role of
the CTI factor in the condition equation. Now, assuming that a
packet is penalized to a lower priority queue based on its positive
E, this penalty will not result in more than expected delay, apart
from potential delay E.
For example, when a packet is inserted queue based on
CT_x <= Q < CT_x + CTI
even if it is at the end of the queue, according to D = Q - E, i.e.,
after time E (the penalty time), we have
CT_x - E <= Q - E < CT_x - E + CTI
That is
CT_y <= D < CT_y + CTI
So, in essence, it is still equivalent to an eligible packet entering
the corresponding queue based on its delay level, and apply the
schedulability condition.
10.1. Packet Disorder Considerations
Suppose that two packets, P1, P2, are generated instantaneously from
a specific flow at the source, and the two packets have the same
planned residence time. P1 may face less interference delay than P2
in their journey. When they arrive at an intermediate node in turn,
P2 will have less allowable queueing delay (Q) than P1 to try to stay
close to P1 again. It should be noted that to compary who is ealier
is based on queue's CT and packet's Q, according to the above
queueing rule (CT <= Q < CT+CTI), and the CT of the queue is not
changed in real-time, but gradually with the decreasing step RTI. It
is possible to get an unexpected comparision result.
As shown in Figure 8, P1 and P2 are two packets belonging to the same
flow. The arrival time when they are received on the scheduler is
shown in the figure. Suppose that CTI is 10us, the decreasing step
Peng, et al. Expires 9 February 2025 [Page 29]
Internet-Draft Deadline Queueing Mechanism August 2024
RTI is 1us, and the transmission time of each packet is 0.01us. Also
suppose that the Q values of two adjacent packets P1 and P2 are 40us
and 39us respectively, and they are both received in the window from
T0 to T0+1us. P1 will enter queue-B with CT range [40, 50), while P2
will enter queue-A with CT range [30, 40) just before the rotation
event occurred. This means that P2 will be scheduled before P1,
resulting in disorder.
packets arrived later
packets arrived earlier |
| |
V V
--------+-----------------------------------------------+---------
... ... | .P1.................P2....................... | ... ...
--------+-----------------------------------------------+---------
P1.Q=40us
P2.Q=39us
| | |
--------o---------------------o---------------------o----------->
T0 T0+1us T0+2us time
queue-A.CT[30,40) queue-A.CT[29,39)
queue-B.CT[40,50) queue-B.CT[39,49)
queue-C.CT[50,60) queue-C.CT[49,59)
Figure 8: Disorder Illustration of RPQ
DetNet architecture [RFC8655] provides Packet Ordering Function
(POF), that can be used to solve the above disorder problem caused by
the latency compensation.
Alternatively, Section 11 provides E|D decoupling method to firstly
absorb latency deviation E by the pre-scheduler which may maintain
FIFO queue per incoming port plus delay level. In this case, packets
from the same flow will only determine the damping delay, but not the
position, in the FIFO based on latency deviation E, to avoid
disorder. Latency deviation E no longer works in the post-scheduler.
Note that in practical situations, two back-to-back packets of the
same flow are generally evenly distributed within the burst interval
by policing, which means that the distance between these two packets
is generally much greater than the calculation accuracy mentioned
above, meaning that the disordered phenomenon will not really occur.
For example, the regulated result meets a Length Rate Quotient (LRQ)
constraint.
Peng, et al. Expires 9 February 2025 [Page 30]
Internet-Draft Deadline Queueing Mechanism August 2024
11. Jitter Performance by On-time Scheduling
The enqueue and dequeue rule of on-time mode described in Section 9
and Section 10 will absorb latency deivation E on each hop, and
archieve a low jitter. The ultimate E2E jitter depends on the delay
experienced on the last node of the flow, which may be from 0 to the
delay bound, i.e., the corresponding delay level d_i.
Depending on different methods of absorbing E, there are slight
differences in scheduling behavior.
* E+D integration: E will be added to D to get the adjusted D'; The
packet is scheduled by the EDF scheduler configured with on-time
mode based on D'.
* E|D decoupling: There are 2-tier schedulers; The packet is
scheduled by pre-scheduler configured with on-time mode based on
E, then scheduled by post-scheduler configured with in-time mode
based on D.
In the case of E+D integration, it may explicitly introduce the
mandatory hold time, and cause that the actual departure time of the
packet may be after its deadline. Asumming that the eligible
arrivals pattern of all delay levels cause the scheduler to work at
full speed (i.e., service rate C), for in-time mode, the worst case
is that there may be a packet of a specific delay level to be sent
just before its deadline during the busy period; While for E+D
integration case, the busy period may just start at its deadline and
cause the sending time of the packet to exceed its deadline.
However, as mentioned above, the worst case of this exceeding value
will not exceed the delay level value, which is intuitive because it
is equivalent to the situation where the observed packet arrives
asynchronously after the delay level value. Note that this exceeding
deadline does not accumulate with the number of hops. The E2E
latency is in the range [D*hops, D*hops+d_i].
In the case of E|D decoupling, the explicitly mandatory hold time is
only contributed by E ensured by the pre-scheduler (configured with
on-time mode), and the actual departure time (from the post-
scheduler) of the packet will always be before its deadline.
Asumming that the eligible arrivals pattern of all delay levels cause
the post-scheduler (configured with in-time mode) to work at full
speed, for E|D decoupling case, the worst case is that there may be a
packet of a specific delay level to be sent just before its deadline
during the busy period. The E2E latency is in the range [D*(hops-1),
D*(hops-1)+d_i]. Note that the pre-scheduler may maintain a PIFO, an
RPQ, or serveral FIFO queues each for particular incoming port plus
delay level. Figure 9 shows the functional entities inside the node.
Peng, et al. Expires 9 February 2025 [Page 31]
Internet-Draft Deadline Queueing Mechanism August 2024
+--------------------------------------------------+
| +---+ +-+ +---------+ +---------+ |
| | | |X| |Pre- | |Post- | |
Input | | S | |X| |Scheduler| |Scheduler| | Output
port -O -> | & | -> |X| -> | (by E) | -> | (by D) | -> O- port
| | F | |X| |[*]PIFO | |[*]PIFO | |
| | | |X| |[*]RPQ | |[*]RPQ | |
| | | |X| |[*]FIFO | | | |
| +---+ +-+ +---------+ +---------+ |
+--------------------------------------------------+
| |
------o-----------------------------------o------------------->
arrive_time_I arrive_time_S time
|<-------- F ------->|<---- E ----->|<---- D-F --->|
Figure 9: E|D decoupling with 2-tier Schedulers
The following Figure 10 shows the difference between on-time
scheduling and in-time scheduling.
Peng, et al. Expires 9 February 2025 [Page 32]
Internet-Draft Deadline Queueing Mechanism August 2024
arrival flows:
A_1: #1 #2 #3 #4 #5 ...
A_2: $1 $2 $3 $4 $5 ...
...
A_5: &1 &2 &3 &4 &5 ...
|
v
In-time Scheduling:
#1$1...&1 #2$2...&2 #3$3...&3 #4$4...&4 #5$5...&5 ...
On-time Scheduling (E+D integration):
#1 #2 #3 #4 #5
$1 $2 $3 $4
... ...
&1
On-time Scheduling (E|D decoupling):
#1$1...&1 #2$2...&2 #3$3...&3 #4$4...&4 #5$5...&5 ...
------+---------+---------+---------+--... ...--+---------+---->
\_ d_1 _/ time
\______ d_2 ______/
\___________ d_3 ___________/
... ...
\_____________________ d_n _______________________/
Figure 10: Difference between In-time and On-time Scheduling
As shown in the figure, each burst of A_1 (corresponding to delay
level d_1) is termed as #num, each burst of A_2 (corresponding to
delay level d_2) as $num, and each burst of A_5 (corresponding to
delay level d_5) as &num. A single burst may contain multiple
packets. For example, burst #1 may contain several packets, and the
actual time interval between #1 and #2 may be small. Although the
figure shows the example that the burst interval of multiple flows is
the same and the phase is aligned, the actual situation is far from
that. However, this example depicts the typical scheduling behavior.
In the in-time scheduling, all concurrent traffic of multiple levels
will be scheduled as soon as possible according to priority, to
construct a busy period. For example, in the duration d_1, in
addition to the burst #1 that must be sent, the burst $1~&1 may also
be sent, but the latter is not necessarily scheduled to be sent
before the burst #2 as shown in the figure. Here we clearly see that
in-time scheduling cannot guarantee jitter.
Peng, et al. Expires 9 February 2025 [Page 33]
Internet-Draft Deadline Queueing Mechanism August 2024
While in the case of on-time scheduling with E+D integration option,
each burst is scheduled at its deadline, which may just be the begin
of the busy period. Because of the scheduling delay, the
transmission of the burst will exceed its deadline. The last packet
of the burst will face more delay than the first packet. For
example, when burst #5 enters the PIFO, it may have the same deadline
with bursts from $4 of A_2 to &1 of A_5. When the deadlines of
multiple packets are the same, use planned residence time (D) as
tiebreaker, i.e., the smaller the D, the smaller the rank. So, #5
send first and may exceed the deadline by one d_1; Then send $4 and
may exceed the deadline by one d_2; ...; Finally, send &1 and may
exceed the deadline by one d_5.
In the case of on-time scheduling with E|D decoupling, assuming that
the latency deviation E for each burst is 0 in the above figure, all
concurrent traffic of multiple levels, similar to in-time, will also
be scheduled as soon as possible according to priority, to construct
a busy period. For example, in the duration d_1, in addition to the
burst #1 that must be sent, the burst $1~&1 may also be sent.
However, $1~&1 will receive punishment based on their E on the next
node (not shown in the figure).
12. Resource Reseravtion
Generally, a path may carry multiple DetNet flows with different
delay levels. For a certain delay level d_i, the path will reserve
some resources from the delay resource pool of the link. The delay
resource pool here, as leaky bucket constraint function shown in
Section 3.2.1 or Section 4.3.2, is a set of preset parameters that
meet the schedulability conditions. For example, the level d_1 has a
burst upper limit of b_1 and a bandwidth upper limit of r_1. A path
j may allocate partial resources (b_i_j, r_i_j) from the resource
pool (b_i, r_i) of the link's delay level d_i. A DetNet flow k that
carried in path j, may use resources (b_i_j_k, r_i_j_k) according to
its T_SPEC. It can be seen that the values of b_i_j and r_i_j
determine the scale of the number of paths that can be supported,
while the values of b_i_j_k and r_i_j_k determine the scale of the
number of flows that can be supported. The following expression
exists.
* sum(b_i_j_k) <= b_i_j, for all flow k over the path j.
* sum(r_i_j_k) <= r_i_j, for all flow k over the path j.
* sum(b_i_j) <= b_i, for all path j through the specific link.
* sum(r_i_j) <= r_i, for all path j through the specific link.
Peng, et al. Expires 9 February 2025 [Page 34]
Internet-Draft Deadline Queueing Mechanism August 2024
12.1. Delay Resource Definition
The delay resources of a link can be represented as the corresponding
burst and bandwidth resources for each delay level. Basically, what
delay levels (e.g., 10us, 20us, 30us, etc) are supported by a link
should be included in the link capability.
Figure 11 shows the delay resource model of the link. The resource
information of each delay level includes the following attributes:
* Delay Bound: Refers to the delay bound intra node corresponding to
this delay level. It is a pre-configuration value.
* Maximum Reservable Bursts: Refers to the maximum amount of bit
quota corresponding to this delay level. It is a pre-allocated
value or resource limit set based on the schedulability condition.
* Utilized Bursts: Refers to the burst utilization of this delay
level.
* Maximum Reservable Bandwidth: Refers to the maximum amount
bandwidth corresponding to this delay level. It is a pre-
allocated value or resource limit set based on the schedulability
condition.
* Utilized Bandwidth: Refers to the bandwidth utilization of this
delay level.
Peng, et al. Expires 9 February 2025 [Page 35]
Internet-Draft Deadline Queueing Mechanism August 2024
d_n +----------------------------------------+
| Maximum Reservable Bursts: MRBur_n |
| Utilized Bursts: UBur_n |
| Maximum Reservable Bandwidth: MRBan_n |
| Utilized Bandwidth: UBan_n |
+----------------------------------------+
... ... ...
... ... ...
d_2 +----------------------------------------+
| Maximum Reservable Bursts: MRBur_2 |
| Utilized Bursts: UBur_2 |
| Maximum Reservable Bandwidth: MRBan_2 |
| Utilized Bandwidth: UBan_2 |
+----------------------------------------+
d_1 +----------------------------------------+
| Maximum Reservable Bursts: MRBur_1 |
| Utilized Bursts: UBur_1 |
| Maximum Reservable Bandwidth: MRBan_1 |
| Utilized Bandwidth: UBan_1 |
+----------------------------------------+
----------------------------------------------------------->
Unidirectional Link
Figure 11: Delay Resource of the Link
For a specific link:
* If Maximum Reservable Bursts and Maximum Reservable Bandwidth are
used to participate schedulability condition checking, they need
to set reasonable values at the beginning to meet the
schedulability condition, and in the future, there is no need to
excecute schedulability condition checking during the setup
procedure of any flow passing through this link, but only need to
check that the aggregated burst and bandwidth of all flows
belonging to the same delay level do not exceed Maximum Reservable
Bursts and Maximum Reservable Bandwidth, respectively.
* If Utilized Bursts and Utilized Bandwidth are used to participate
schedulability condition checking, there is necessary to excecute
schedulability condition checking during the setup procedure of
any new flows passing through this link.
The IGP/BGP extensions to advertise the link's capability and delay
resource is defined in
[I-D.peng-lsr-deterministic-traffic-engineering].
Peng, et al. Expires 9 February 2025 [Page 36]
Internet-Draft Deadline Queueing Mechanism August 2024
12.2. Traffic Engineering Path Calculation
A candidate path may be selected according to the end-to-end delay
requirement of the flow. Subtract the accumulated link propagation
delay from the end-to-end delay requirement, and then divide it by
the number of hops to obtain the average planned residence time (D)
for each node. Or, different nodes may have different planned
residence time (D). By default, select the appropriate delay level
d_i (d_i <= D-F) closest to the planned residence time (D), and then
reserve resources from delay level d_i on each hop. A local policy
may allow more larger D to consume resources with smaller delay
levels.
Note that it is planned residence time (D), not delay level (d_i),
carried in the forwarding packets.
13. Policing on the Ingress
On the ingress PE node, policing must be performed on the incoming
port, so that DetNet flow does not exceed its T-SPEC. This kind of
traffic regulation is usually the shaping using leaky bucket. After
policing, the shaped pattern of the DetNet flow may contain discrete
multiple bursts evenly distributed within its periodic service burst
interval (SBI). For example, An arriving elephant flow will be
diluted and released to the EDF scheduler.
According to [RFC9016], the values of Burst Interval,
MaxPacketsPerInterval, MaxPayloadSize of the DetNet flow will be
written in the SLA between the customer and the network provider, and
the network entry node will set the corresponding bucket depth
according to MaxPayloadSize to forcibly delay the excess bursts. The
entry node also sets the corresponding bucket rate according to the
promised arrival rate.
The shaped pattern is generally inconsistent with the original
arrival pattern of the DetNet flow, and some bursts of the original
arrival pattern may experience more shaping delay than others. The
shaped pattern and the original arrival pattern can be as consistent
as possible by increasing the bucket depth, but this means that the
flow will occupy more burst resources, and reduce the service scale
that the network can support according to the schedulability
conditions.
Peng, et al. Expires 9 February 2025 [Page 37]
Internet-Draft Deadline Queueing Mechanism August 2024
On the network entrance node, for the burst with applied shaping
delay, shaping delay cannot be included in the latency compensation
equation, otherwise, it will make that burst catch up with the
previous burst, resulting in damage to the policing result and
violation of the arrival constraint function. Please refer to
[I-D.peng-detnet-policing-jitter-control] for the elimination of
jitter caused by shaping delay on the network entrance node.
Then, the regulated traffic arrives at the EDF scheduler on the
outgoing port. Since the traffic of each delay level meets the leaky
bucket arrival constraint function and the parameters of the shaping
curve do not exceed the limits of the parameters provided by the
schedulability conditions, the traffic can be successfully scheduled
based on deadline.
Note that the flow arrived at the next hop, after reshaping or
latency compensation, will still follow the arrival constraint
function of that flow. When this flow is aggregated with other flows
and sent to the same outgoing port, within any duration d_i, the
aggregated d_i traffic will not exceed the burst and bandwidth
resources of delay level d_i reserved by these flows on the outgoing
port.
Figure 12 depicts an example of policing and deadline based
scheduling on the ingress PE node in the case of option-4 with on-
time mode. In the figure, the shaping delay caused by the previous
burst is termed as S#.
Peng, et al. Expires 9 February 2025 [Page 38]
Internet-Draft Deadline Queueing Mechanism August 2024
1st burst
|
received v
+-+ +-+ +----+ +-+ +--+ +------+
|1| |2| | 3 | |4| |5 | | 6 | <= burst
+-+ +-+ +----+ +-+ +--+ +------+ sequence
| | | | | |
~+0 ~+S1 ~+0 ~+S3~+S4 ~+0 <= shaping
| \ | | \_ | delay
| | | | | |
UNI v v v v v v
ingr-PE -+--------+--------+--------+--------+--------+--------+---->
NNI | CTI | CTI | CTI | CTI | CTI | CTI | time
1,2 in 3 in 4 in 5 in 6 in
queue-A queue-B queue-C queue-D queue-E
A.CT<=Q B.CT<=Q C.CT<=Q D.CT<=Q E.CT<=Q
| | | | |
~+Q ~+Q ~+Q ~+Q ~+Q <= e.g., Q = CTI
\_____ \_____ \_____ \_____ \_____
| | | | |
sending v(que-A) v(que-B) v(que-C) v(que-D) v(que-E)
+-+-+ +----+ +-+ +--+ +------+
|1|2| | 3 | |4| |5 | | 6 |
+-+-+ +----+ +-+ +--+ +------+
Figure 12: Deadline Based Packets Orchestrating
There are 6 bursts received from the client. The burst-2, 4, 5 has
policing delay S1, S3, S4 that caused by previous burst respectively.
While burst-1, 3, 6 has zero policing delay because the number of
tokens is sufficient. The policing makes 6 bursts roughly
distributed within the service burst interval.
Assuming that the forwarding delay F experienced by all bursts is 0.
In the case of latency compensation plus RPQ, they will have the same
allowable queueing delay (Q), regardless of whether they have
experienced policing delay before. When the packets of burst-1, 2
arrive at the scheduler, according to CT <= Q < CT+CTI, they will be
placed in Queue-A with matched CT and waiting to be sent. Similarly,
when the packets of burst-3/4/5/6 arrive at the scheduler, they will
be placed in Queue-B/C/D/E respectively and waiting to be sent
according to the de-queue rules of on-time mode. Note that each
sending burst may get a latency deviation E, especially for burst-2,
which is sent closely adjacent to burst-1 in the sending pattern.
Peng, et al. Expires 9 February 2025 [Page 39]
Internet-Draft Deadline Queueing Mechanism August 2024
14. Overprovision Analysis
For each delay level d_i, the delay resource of the specific link is
(b_i, r_i). A path j may allocate partial resources (b_i_j, r_i_j)
from the resource pool (b_i, r_i). In order to support more d_i
flows in the network, it is necessary to set larger b_i and r_i.
However, as mentioned earlier, the values of b_i and r_i are set
according to schedulability conditions and cannot be set at will.
Therefore, the meaningful analysis is the service scale that the
network can support under the premise of determined b_i and r_i.
For bandwidth resource reservation case, the upper limit of the total
bandwidth that can be reserved for all aggregated flows of delay
level d_i is r_i, which is the same as the behavior of traditional
bandwidth resource reservation. There is no special requirement for
the measurement interval of calculating bandwidth value.
For the burst resource reservation case, the upper limit of the total
burst that can be reserved for all aggregated flows of delay level
d_i is b_i. If the burst of each flow of level d_i is b_k, then the
number of flows can be supported is b_i/b_k, which is the worst case
considering the concurrent arrival of these flows. However, the
burst resource reservation is independent of bandwidth resource,
i.e., it does not take the calculation result of b_k/d_i to get an
overprovision bandwidth and then to affect the reservable bandwidth
resources.
By providing multiple delay levels, we can allocate 100% of the link
bandwidth to DetNet flows, as can be seen from the schedulability
condition equation.
15. Compatibility Considerations
Deadline is suitable for end-to-end and interconnection between
different networks. A large-scale network may span multiple
networks, and one of the goals of DetNet is to connect each network
domain to provide end-to-end deterministic delay service. The
adoption techniques and capabilities of each network are different,
and the corresponding topology models are either piecewise or nested.
For a particular path, if only some nodes in the path upgrade support
the deadline based mechanism defined in this document, the end-to-end
deterministic delay/jitter target will only be partially achieved.
Those legacy devices may adopt the existing SP or WFQ mechanisms, and
ignore the possible deadline information carried in the packet, thus
the residence delay produced by them cannot be perceived by the
adjacent upgraded node. The more upgraded nodes included in the
path, the closer to the delay/jitter target. Although, the legacy
Peng, et al. Expires 9 February 2025 [Page 40]
Internet-Draft Deadline Queueing Mechanism August 2024
devices may not support the data plane mechanism described in this
document, but they can be freely programmed (such as P4 language) to
measure and insert the deadline information into packets, in this
case the delay/jitter target may be achieved.
Only a few key nodes are upgraded to support deadline mechanism,
which is low-cost, but can meet a flow with relatively loose time
sensitive. Figure 13 shows an example of upgrading only several
network border nodes. In the figure, only R1, R2, R3 and R4 are
upgraded to support deadline based mechanism. A deterministic path
across domain 1, 2, and 3 is established, which contains nodes R1,
R2, R3, and R4, as well as explicit nodes in each domain. Domain 1,
2 and 3 use the traditional SP mechanism. The encoding of the packet
sent by R1 includes the planned residence time and the latency
deviation E. Especially, DS filed in IP header ([RFC2474]) are also
set to appropriate values. The basic principle of setting is that
the less the planned residence time, the higher the priority. In
order to avoid the interference of non deterministic flow to
deterministic flow, the priority of deterministic flow should be set
as high as possible.
The delay analysis based on strict priority without re-shaping in
each domain can be found in [SP-LATENCY], which gives the equation to
evaluate the worst-case delay of each hop. The worst-case delay per
hop depends on the number of hops and the burst size of interference
flows that may be faced on each hop. [EF-FIFO] also shows that, for
FIFO packet scheduling be used to support the EF (expedited
forwarding) per-hop behavior (PHB), if the network utilization level
alpha < l/(H-l), the worst-case delay bound is inversely proportional
to 1-alpha*(H-1), where H is the number of hops in the longest path
of the network.
Although the EDF scheduling with in-time mode, the SP scheduling and
EF FIFO scheduling are all work-conserving, the EDF scheduling can
further distinguish between urgent and non urgent packets according
to deadline information other than traffic class. An intuitive
phenomenon is that if a packet unfortunately faces more interference
delays at the upstream nodes, it will become more urgent at the
downstream node, and will not always be unfortunate. This operation
of dynamically modifying the key fields, i.e., the latency deviation
(E), of the packet can avoid always overestimating worst-case latency
on all hops just like SP.
For a specific DetNet flow, if it experiences too much latency in the
SP domain (due to unreasonable setting of DS field and the inability
to distinguish between deterministic and non deterministic flows),
even if the boundary node accelerates the transmission, it may not be
able to achieve the target of low E2E latency strictly, but rather a
Peng, et al. Expires 9 February 2025 [Page 41]
Internet-Draft Deadline Queueing Mechanism August 2024
loose result. If the traffic experiences less latency within the SP
domain, the on-time scheduling mode applied on the border node can
help achieve the end-to-end jitter target.
_____ ___ _____ ___ _____ ___
/ \/ \___ / \/ \___ / \/ \___
/ \ / \ / \
+--+ +--+ +--+ +--+
|R1| Strict Priority |R2| Strict Priority |R3| Strict Priority |R4|
+--+ domian 1 +--+ domian 2 +--+ domian 3 +--+
\____ __/ \____ __/ \____ __/
\_______/ \_______/ \_______/
Figure 13: Example of partial upgrade
16. Deployment Considerations
According to the above schedulability conditions, each delay level
d_i has dedicated delay resources, and the smaller d_i, the more
valuable it is. The operator needs to match the corresponding d_i
for each flow. It should be noted that the per-hop latency provided
by the deadline based mechanism for the flow is based on flow's
RSpec, not TSpec.
In the case of option-3 and 4 with in-time scheduling behavior, more
buffer is required to store accumulated bursts.
For a specific flow, the accumulated bursts on a intermediate node
consists of multiple rounds of burst interval. For example, the
packets generated by the source within the first round of burst
interval (always experiencing the worst case delay along the path) is
caught up by the packets generated within the second round of burst
interval (always experiencing the best case delay along the path).
For delay level d_i, the worst case delay is d_i, the best case delay
is l/R, where l is the smallest packet size of the flow, R is the
port rate. For simplicity to get the estimate size of accumulated
bursts, here we just take the best case delay as 0. Drawing on the
method provided in [SP-LATENCY], the accumulated bursts of d_i is:
* ACC_BUR_i = ((d_i * h) / burst_interval) * b_i
Peng, et al. Expires 9 February 2025 [Page 42]
Internet-Draft Deadline Queueing Mechanism August 2024
For example, d_i is 10 us, burst_interval is 250 us, this means that
within the 25th hop, there will only be one b_10 burst in the queue.
If it exceeds 25 hops and is within 50 hops, there may be two b_10
burst simutaneously in the queue.
The accumulated bursts of other delay levels can be similarly
estimated. Operators need to evaluate the required buffer size based
on network hops and the supported delay levels. The benefit of in-
time scheduling is to obtain an E2E latency of no more than D*hops as
small as possible.
Operators may also apply on-time scheduling to simplify the design of
buffers. On-time scheduling absorbed latency deviation E on each hop
and can get a jitter for each delay level to the value of delay level
in theory (i.e., the worst case is that on the last node there are
full traffic contributed by all delay levels that are discharging
floodwater at the same time, however, in reality, the DetNet flow of
the output port facing the destination customer side may only involve
one delay level, then the jitter may be only one CTI (e.g., 10us)).
In summary, the in-time scheduling with latency compensation, can
suffer from the uncertainty caused by burst accumulation, and it is
recommended only deployed in small networks, i.e., a limited domain
with a small number of hops, where the burst accumulation issue is
not serious; The on-time scheduling is recommended to be used in
large networks.
17. Evaluations
Now we summarize how the deadline based mechanism ensures bounded
latency as below:
1) Partition delay resource for each delay level on the outgoing
port according to the schedulability condition, i.e., preset
parameters of the arrival constraint function.
2) Execute delay resource reservation on each port of each
calculated path on the control plane. This step is also known as
admission condition check.
3) Execute policing on the network entry, to let the admited
traffic obey its constraint function (i.e., TSpec).
4) Execute reshaping or latency compensation (recommended) in the
network core for each flow, to convert the ineligible arrivals to
eligible arrivals that still obey the constraint function of each
flow.
Peng, et al. Expires 9 February 2025 [Page 43]
Internet-Draft Deadline Queueing Mechanism August 2024
5) Guarantee bounded delay by in-time scheduling mode; Guarantee
bounded delay and jitter by on-time scheduling mode.
The following table is the evaluation results of the Deadline
mechanism based on the requirements that is defined in
[I-D.ietf-detnet-scaling-requirements]. Note that all asynchronous
mechanisms (such as EDF, ATS) do not require complete synchronization
of crystal oscillator frequencies between devices, and the latency
error caused by the deviations of clocks from their nominal rates
(e.g., +100ppm) is generally in the nanosecond range and can be
ignored.
+======================+============+===============================+
| requiremens | Evaluation | Notes |
+======================+============+===============================+
| 3.1 Tolerate Time | Yes | No time sync needed; |
| Asynchrony | | No frequency sync needed. |
+----------------------+------------+-------------------------------+
| 3.2 Support Large | | The eligible arrival of |
| Single-hop | Yes | flows is independent with the |
| Propagation | | link propagation delay. |
| Latency | | |
+----------------------+------------+-------------------------------+
| 3.3 Accommodate the | | The higher service rate, the |
| Higher Link | Partial | more buffer needed for each |
| Speed | | delay level. And, extra |
| | | instructions to calculate E. |
+----------------------+------------+-------------------------------+
| 3.4 Be Scalable to | | Multiple delay levels, each |
| the Large Number | | with limited delay resources, |
| of Flows and | | can support lots of flows, |
| Tolerate High | Yes | without overprovision. |
| Utilization | | Utilization may reach 100% |
| | | link bandwidth. |
| | | The unused bandwidth of the |
| | | high delay level can be used |
| | | by the low levels or BE flows.|
+----------------------+------------+-------------------------------+
| 3.5 Tolerate Failures| | Independent of queueing |
| of Links or Nodes| N/A | mechanism. |
| and Topology | | |
| Changes | | |
+----------------------+------------+-------------------------------+
| 3.6 Prevent Flow | | Flows are permitted based on |
| Fluctuation | Yes | the resources reservation of |
| | | delay levels, and isolated |
| | | from each other. |
Peng, et al. Expires 9 February 2025 [Page 44]
Internet-Draft Deadline Queueing Mechanism August 2024
+----------------------+------------+-------------------------------+
| 3.7 Be scalable to a | | E2E latency is liner with hops|
| Large Number of | | , from ultra-low to low |
| Hops with Complex| Yes | latency by multiple delay |
| Topology | | levels. |
| | | E2E jitter is low by on-time |
| | | scheduling. |
+----------------------+------------+-------------------------------+
| 3.8 Support Multi- | | Independent of queueing |
| Mechanisms in | N/A | mechanism. |
| Single Domain and| | |
| Multi-Domains | | |
+----------------------+------------+-------------------------------+
Figure 14: Evaluation for Large Scaling Requirements
17.1. Examples
This section describes the example of how the deadline mechanism
supports DetNet flows with different latency requirements.
17.1.1. Heavyweight Loading Example
This example observes the service scale and different latency bound
supported by the deadline mechanism in the heavyweight loading case.
Figure 15 provides a typical reference topology that serves to
represent or measure the multihop jitter and latency experience of a
single "flow i" across N hops (in the figure, N=10). On each of the
N outgoing interfaces (represented by circles in the figure), "flow
i" has to compete with different flows (represented by different
symbols on each hop). Note that this topology could most likely
represent a part of a typical service provider access ring topology,
such as in mobile network backhaul in rural areas. The
characteristic of this reference topology is that every link that
"flow i" passes through may be a bottleneck link with 100% network
utilization, causing "flow i" to achieve the worst-case latency on
each hop.
As shown in Figure 15:
* Network transmission capacity: each link has rate 10 Gbps.
Assuming the service rate of EDF scheduler allocate the total port
bandwidth.
* TSpec of each flow, maybe:
Peng, et al. Expires 9 February 2025 [Page 45]
Internet-Draft Deadline Queueing Mechanism August 2024
- burst size 1000 bits, and average arrival rate 1 Mbps.
- or, burst size 1000 bits, and average arrival rate 10 Mbps.
- or, burst size 1000 bits, and average arrival rate 100 Mbps.
- or, burst size 10000 bits, and average arrival rate 1 Mbps.
- or, burst size 10000 bits, and average arrival rate 10 Mbps.
- or, burst size 10000 bits, and average arrival rate 100 Mbps.
* RSpec of each flow, maybe:
- E2E latency 100us, and E2E jitter less than 10us or 100us.
- or, E2E latency 200us, and E2E jitter less than 20us or 200us.
- or, E2E latency 300us, and E2E jitter less than 30us or 300us.
- or, E2E latency 400us, and E2E jitter less than 40us or 400us.
- or, E2E latency 500us, and E2E jitter less than 50us or 500us.
- or, E2E latency 600us, and E2E jitter less than 60us or 600us.
- or, E2E latency 700us, and E2E jitter less than 70us or 700us.
- or, E2E latency 800us, and E2E jitter less than 80us or 800us.
- or, E2E latency 900us, and E2E jitter less than 90us or 900us.
- or, E2E latency 1ms, and E2E jitter less than 100us or 1ms.
Peng, et al. Expires 9 February 2025 [Page 46]
Internet-Draft Deadline Queueing Mechanism August 2024
@ # $
@ # $
v v v
+---+ @@@ +---+ ### +---+ $$$ &&& +---+
src ---> | 0 o-----| 1 o-----| 2 o---- ... ... ----| 9 o----> dest
(flow i:*)+---+ *** +---+ *** +---+ *** *** +---+ ***
| |@ |# |&
| |@ |# |&
| |v |v |v
+---+ +---+ +---+ +---+
--- | | --- | | --- | | --- ... ... --- | | ---
+---+ +---+ +---+ +---+
| | | |
| | | |
... ... ... ... ... ... ... ...
Figure 15: Heavyweight Loading Topology Example
For the observed flow i (marked with *), its TSpec and RSpec may be
any of the above. Assuming that the path calculated by the
controller for the flow i passes through 10 nodes (i.e., node 0~9).
Especially, at each hop, flow i may conflict with other deterministic
flows, also with similar TSpec and RSpec as above, originated from
other sources, e.g., conflicts with flow-set "@" at node 0, conflicts
with flow-set "#" at node 1, and so on.
For each link along the path, it may provide delay resources with
multiple delay levels, e.g., d1 (10us), d2 (20us), ..., d10 (100us)
for any flows passing through it to consume. Assuming no link
propagation delay and intra node forwarding delay, if flow i uses d1
resources, it can ensure an E2E latency of 100us (i.e., d1 * 10
hops), and jitter of 10us(on-time mode) or 100us (in-time mode). The
results of using resources of other delay levels are similar.
The table below shows the possible tight allocation of delay
resources on each link based on Equation-1, as well as the
corresponding service scale supported, where, b = utilized burst
resource (K bits), r = utilized bandwidth resource (Mbps), s =
service scale (number), assuming that the resource limit of each
delay level is b_limit = 100000 bits, r_limit = 1 Gbps.
Note that in the table each column only shows the data where all
flows served by all delay levels have the same TSpec (e.g., in
colunm-1, TSpec per flow is burst size 1000 bits and arrival rate 1
Mbps), while in reality, flows served by different delay levels
generally have different TSpec. It is easy to add colunms to
describe various combinations.
Peng, et al. Expires 9 February 2025 [Page 47]
Internet-Draft Deadline Queueing Mechanism August 2024
===================================================
| d1 | d2 | d3 | d4 | d5 | d6 | d7 | d8 | d9 | d10|
===================================================================
| column-1 | b | 100| 99 | 98 | 97 | 96 | 95 | 94 | 93 | 92 | 91 |
| |---+----+----+----+----+----+----+----+----+----+----|
|TSpec: | r | 100| 99 | 98 | 97 | 96 | 95 | 94 | 93 | 92 | 91 |
| 1000 bits |---+----+----+----+----+----+----+----+----+----+----|
| 1 Mbps | s | 100| 99 | 98 | 97 | 96 | 95 | 94 | 93 | 92 | 91 |
===================================================================
| column-2 | b | 100| 90 | 81 | 73 | 66 | 60 | 53 | 48 | 43 | 39 |
| |---+----+----+----+----+----+----+----+----+----+----|
|TSpec: | r |1000| 900| 810| 729| 656| 590| 531| 478| 430| 387|
| 1000 bits |---+----+----+----+----+----+----+----+----+----+----|
| 10 Mbps | s | 100| 90 | 81 | 72 | 65 | 59 | 53 | 47 | 43 | 38 |
===================================================================
| column-3 | b | 100| 90 | 80 | 70 | 60 | 50 | 40 | 30 | 20 | 10 |
| |---+----+----+----+----+----+----+----+----+----+----|
|TSpec: | r |1000|1000|1000|1000|1000|1000|1000|1000|1000|1000|
| 1000 bits |---+----+----+----+----+----+----+----+----+----+----|
| 100 Mbps | s | 10 | 10 | 10 | 10 | 10 | 10 | 10 | 10 | 10 | 10 |
===================================================================
| column-4 | b | 100| 100| 100| 100| 100| 100| 99 | 99 | 99 | 99 |
| |---+----+----+----+----+----+----+----+----+----+----|
|TSpec: | r | 10 | 9 | 9 | 9 | 9 | 9 | 9 | 9 | 9 | 9 |
| 10000 bits|---+----+----+----+----+----+----+----+----+----+----|
| 1 Mbps | s | 10 | 9 | 9 | 9 | 9 | 9 | 9 | 9 | 9 | 9 |
===================================================================
| column-5 | b | 100| 99 | 98 | 97 | 96 | 95 | 94 | 93 | 92 | 91 |
| |---+----+----+----+----+----+----+----+----+----+----|
|TSpec: | r | 100| 99 | 98 | 97 | 96 | 95 | 94 | 93 | 92 | 91 |
| 10000 bits|---+----+----+----+----+----+----+----+----+----+----|
| 10 Mbps | s | 10 | 9 | 9 | 9 | 9 | 9 | 9 | 9 | 9 | 9 |
===================================================================
| column-6 | b | 100| 90 | 81 | 73 | 66 | 59 | 53 | 48 | 43 | 39 |
| |---+----+----+----+----+----+----+----+----+----+----|
|TSpec: | r |1000| 900| 810| 729| 656| 590| 531| 478| 430| 387|
| 10000 bits|---+----+----+----+----+----+----+----+----+----+----|
| 100 Mbps | s | 10 | 9 | 8 | 7 | 6 | 5 | 5 | 4 | 4 | 3 |
===================================================================
Figure 16: Delay Resource Pool and Service Scale Example
17.1.2. Lightweight Loading Example
This example observes how the preset service scale is supported and
mapped to different delay levels by the deadline mechanism in the
lightweight loading case.
Peng, et al. Expires 9 February 2025 [Page 48]
Internet-Draft Deadline Queueing Mechanism August 2024
[I-D.ietf-detnet-dataplane-taxonomy] describes a grid topology
(Figure 17) with partial mesh. Three flow types, i.e., audio, vedio,
and CC (Command and Control) are considered to require deterministic
networking services. Among them, audio and CC flows consume less
bandwidth (1.6 Mbps per flow and 0.48 Mbps per flow respectively) but
both require lower E2E latency (5ms), while vedio flows consume more
bandwidth (11 Mbps per flow) but can tolerate larger E2E latency
(10ms).
Src 1 Src 2 Src 3
| | |
+------+ +------+ +------+
Dst 1-|Node 1|<--|Node 2|-->|Node 3|-Dst 4
+------+ +------+ +------+
| ^ |
V | V
+------+ +------+ +------+
Dst 2-|Node 4|-->|Node 5|<--|Node 6|-Dst 5
+------+ +------+ +------+
^ | ^
| V |
+------+ +------+ +------+
Dst 3-|Node 7|<--|Node 8|-->|Node 9|-Dst 6
+------+ +------+ +------+
| | |
Src 4 Src 5 Src 6
Figure 17: Lightweight Loading Topology Example
According to the preset rules that generate a unique route for every
source and destination pair, the details of all paths are as follows:
Src1-1-Dst1
Src1-1-4-Dst2
Src1-1-4-5-8-7-Dst3
Src1-1-4-5-2-3-Dst4
Src1-1-4-5-8-9-6-Dst5
Src1-1-4-5-8-9-Dst6
Src2-2-1-Dst1
Src2-2-1-4-Dst2
Src2-2-3-6-5-8-7-Dst3
Src2-2-3-Dst4
Src2-2-3-6-Dst5
Src2-2-3-6-5-8-9-Dst6
Peng, et al. Expires 9 February 2025 [Page 49]
Internet-Draft Deadline Queueing Mechanism August 2024
Src3-3-6-5-2-1-Dst1
Src3-3-6-5-2-1-4-Dst2
Src3-3-6-5-8-7-Dst3
Src3-3-Dst4
Src3-3-6-Dst5
Src3-3-6-5-8-9-Dst6
Src4-7-4-5-2-1-Dst1
Src4-7-4-Dst2
Src4-7-Dst3
Src4-7-4-5-2-3-Dst4
Src4-7-4-5-8-9-6-Dst5
Src4-7-4-5-8-9-Dst6
Src5-8-7-4-5-2-1-Dst1
Src5-8-7-4-Dst2
Src5-8-7-Dst3
Src5-8-7-4-5-2-3-Dst4
Src5-8-9-6-Dst5
Src5-8-9-Dst6
Src6-9-6-5-2-1-Dst1
Src6-9-6-5-2-1-4-Dst2
Src6-9-6-5-8-7-Dst3
Src6-9-6-5-2-3-Dst4
Src6-9-6-Dst5
Src6-9-Dst6
Where, flows to destination Dst1 and Dst6 are audio flows, flows to
destination Dst2 and Dst5 are CC flows, and flows to destination Dst3
and Dst4 are vedio flows. Each path carries 10 flows, e.g., the path
"Src1-1-Dst1" carries 10 audio flows. It can be seen that the
longest path contains 7 hops, and the bottleneck link involves link
(2-3) and link (8-7), both of which have 10 audio flows, 60 video
flows, and 10 CC flows.
Grid topology in this example only contains a small number of
bottleneck links with low network utilization, and it can be
considered as the lightweight loading case of Figure 15. Lightweight
loading usually means having a smaller calculated worst-case latency
per hop, or the actual latency experienced cannot reach the worst-
case latency.
Peng, et al. Expires 9 February 2025 [Page 50]
Internet-Draft Deadline Queueing Mechanism August 2024
According to the longest path and the expected E2E latency, the per-
hop latency bound for each type of flow can be estimated, i.e., 700us
for audio and CC flows, 1400us for vedio flows. This means that the
deadline mechanism needs to provide appropriate delay levels, and the
delay level mapped by audio and CC flows cannot be larger than 700us,
and the delay level mapped by vedio flows cannot be larger than
1400us.
For simplicity, a unified delay resource pool is configured on each
link in the network, although different links can indeed be
configured differently. This unified delay resource pool must meet
the resource allocation requirements on both bottleneck and non
bottleneck links, so we slightly increase the loading and assume that
the number of each type of flows on a link reaches 60. Figure 18
shows a possible delay resource pool and the corresponding delay
levels mapped by flows. Note that there are other possible resource
pool designs as long as they meet schedulability conditions.
===================================================================
| Delay Levels | Bursts | Bandwidth |Services Mapped|
+--------------+-----------------+----------------+---------------+
| d1 (100us) | b1 = 40 Kbits | r1 = 10 Mbps | |
+--------------+-----------------+----------------+---------------+
| d2 (200us) | b2 = 144 Kbits | r2 = 30 Mbps | CC |
+--------------+-----------------+----------------+---------------+
| d3 (300us) | b3 = 0 | r3 = 0 | |
+--------------+-----------------+----------------+---------------+
| d4 (400us) | b4 = 0 | r4 = 0 | |
+--------------+-----------------+----------------+---------------+
| d5 (500us) | b5 = 0 | r5 = 0 | |
+--------------+-----------------+----------------+---------------+
| d6 (600us) | b6 = 0 | r6 = 0 | |
+--------------+-----------------+----------------+---------------+
| d7 (700us) | b7 = 120 Kbits | r7 = 96 Mbps | Audio |
+--------------+-----------------+----------------+---------------+
| d8 (800us) | b8 = 0 | r8 = 0 | |
+--------------+-----------------+----------------+---------------+
| d9 (900us) | b9 = 0 | r9 = 0 | |
+--------------+-----------------+----------------+---------------+
| d10 (1000us) | b10 = 0 | r10 = 0 | |
+--------------+-----------------+----------------+---------------+
| d11 (1100us) | b11 = 720 Kbits | r11 = 660 Mbps | Vedio |
+--------------+-----------------+----------------+---------------+
Figure 18: Delay Resource Pool and Service Mapped
Peng, et al. Expires 9 February 2025 [Page 51]
Internet-Draft Deadline Queueing Mechanism August 2024
Where, the granularity of delay level is choosed at the level of 100
us based on the link capability (1 Gbps) and the concurrent bursts
(984 Kbits) of three type of flows. Intuitively, if the link
capability is larger, such as 10 Gbps, the granularity can be chosen
to be smaller, such as at the level of 10us.
The maximum delay level d11 (1100 us) is selected according to the
minimum regulated packet interval of any flow, i.e., d11 is not
larger than any regulated packet interval of any flow. In this
example, the regulated packet interval (i.e., packet_size /
service_rate) for flows audio, vedio, and CC are 1.25 ms, 1.1 ms, and
5 ms, respectively.
All delay levels consume approximately 800 Mbps bandwidth due to
slightly increasing the loading. 60 CC flows are mapped to delay
level d2, 60 audio flows are mapped to delay level d7, and 60 vedio
flows are mapped to delay level d11. The delay level d1 may be used
for more urgent flows other than the three type of flows considered.
For example, on the bottleneck link (2-3), 10 audio flows will
allocate <burst = 20 Kbits, bandwidth = 16 Mbps> from d7, 60 vedio
flows will allocate <burst = 720 Kbits, bandwidth = 660 Mbps> from
d11, and 10 CC flows will allocate <burst = 24 Kbits, bandwidth = 5
Mbps> from d2.
For example on the non bottleneck link (8-9), 50 audio flows will
allocate <burst = 100 Kbits, bandwidth = 80 Mbps> from d7, zero vedio
flows will not allocate resources from d11, and 30 CC flows will
allocate <burst = 72 Kbits, bandwidth = 15 Mbps> from d2.
If on-time mode is applied, each packet of the audio flow may
experience per-hop latency 700 us, and ecah packet of the CC flow may
experience per-hop latency 200 us, and ecah packet of the vedio flow
may experience per-hop latency 1100 us.
If in-time mode is applied, the best per-hop latency experienced by a
packet in any flow may be 0 (without considering intra-node
forwarding delay F), and the theoretical worst-case latency may be
the same as that in the on-time mode in the case of heavyweight
loading. However, due to lightweight loading in this example,
smaller worst-case latency can be achieved. For example, on the
bottleneck link (2-3), the admitted burst aggregation is composed of
CC 24 Kbits, audio 20 kbits, and vedio 720 Kbits in descending order
of transmission priority, therefore, the worst-case per-hop latency
experienced by the last packet of flows CC, audio, and vedio, is 24
us, 44 us, and 764 us, respectively, which are much smaller than the
values in the on-time mode. Similarly, on the non bottleneck link
(8-9), the admitted burst aggregation is composed of CC 72 Kbits and
Peng, et al. Expires 9 February 2025 [Page 52]
Internet-Draft Deadline Queueing Mechanism August 2024
audio 100 kbits, in descending order of transmission priority,
therefore, the worst-case per-hop latency experienced by the last
packet of flows CC and audio is 72 us and 172 us respectively, which
are also much smaller than the values in the on-time mode.
NOTE:
* In the above process of resource allocation, the 10 flows carried
on each path are individually allocated burst resources. This is
the most general case, that is, although the 10 flows share the
same path, they are assumed to be independent of each other.
However, in some cases, if these 10 flows are treated as a macro
flow and policing is executed at the network entrance node for the
macro flow (the leaky bucket depth is still the maximum packet
size, but the leack bucket rate is the aggregation rate), and
resources are reserved for the macro flow instead of the member
flow, then less burst resources will be consumed and larger
service scales can be supported.
* This example conforms to the scenarios described in Section 3.2.1
and Section 4.3.2 for the application of simplified shcedulability
condition where d_n is not larger than any regulated packet
interval of any flow. Therefore, there are remaining 76 Kbits
bursts available for any delay level d_i, to support more flows.
18. Taxonomy Considerations
[I-D.ietf-detnet-dataplane-taxonomy] provides criteria for
classifying data plane solutions. CEDF is a non-periodic,
asynchronous, class level, work-conserving/non-work-conserving
configurable, in-time/on-time configurable, delay based solution.
* Non-periodic: The scheduling power of an EDF is measured over an
arbitrarily long non repetitive time range, scheduling in an
orderly manner according to the urgency of the packets, and there
is no defined periodic quantification unit of scheduling power.
* Asynchronous: The EDF scheduler always assumes that all DetNet
flows arrive asynchronously and does not require these flows to be
interleaved with each other. The EDF schedulers in the network do
not need to synchronize their states (e.g., busy periold) with
each other, but work independently.
* Class level: DetNet Flows may be grouped by similar service
requirements, i.e., delay levels, on the network entrance node.
Packets will be provided EDF service based on delay level, without
checking flow characteristic.
Peng, et al. Expires 9 February 2025 [Page 53]
Internet-Draft Deadline Queueing Mechanism August 2024
* Work-conserving/non-work-conserving configurable: The EDF
scheduler configured with in-time scheduling mode is work-
conserving (i.e., to send packets as soon as possible), while the
EDF scheduler configured with on-time scheduling mode is non work-
conserving (i.e., to ensure that the packet always experiences the
worst-case delay per hop).
* In-time/on-time configurable: The EDF scheduler configured with
in-time scheduling mode is in-time to get bounded end-to-end
latency, while the EDF scheduler configured with on-time
scheduling mode is on-time to get bounded end-to-end delay jitter.
* Delay based: A DetNet flow is scheduled based on its expected
delay level, but not on its reserved bandwidth (i.e., rate), to
fit well the case of low bandwidth but urgent flows.
In addition, the per hop latency dominant factor of CEDF is simply
delay levels that is defined according to schedulability condition.
19. IANA Considerations
There is no IANA requestion for this document.
20. Security Considerations
Security considerations for DetNet are described in detail in
[RFC9055]. General security considerations for the DetNet
architecture are described in [RFC8655]. Considerations specific to
the DetNet data plane are summarized in [RFC8938].
Adequate admission control policies should be configured in the edge
of the DetNet domain to control access to specific delay resources.
Access to classification and mapping tables must be controlled to
prevent misbehaviors, e.g., an unauthorized entity may modify the
table to map traffic to an expensive delay resource, and competes and
interferes with normal traffic.
21. Acknowledgements
TBD
22. References
22.1. Normative References
[I-D.hp-detnet-tsn-queuing-mechanisms-evaluation]
Yan, J., Han, Y., Peng, S., and Y. Gao, "Analysis and
Evaluation for TSN Queuing Mechanisms", Work in Progress,
Peng, et al. Expires 9 February 2025 [Page 54]
Internet-Draft Deadline Queueing Mechanism August 2024
Internet-Draft, draft-hp-detnet-tsn-queuing-mechanisms-
evaluation-01, 20 December 2023,
<https://datatracker.ietf.org/doc/html/draft-hp-detnet-
tsn-queuing-mechanisms-evaluation-01>.
[I-D.ietf-detnet-dataplane-taxonomy]
Joung, J., Geng, X., Peng, S., and T. T. Eckert,
"Dataplane Enhancement Taxonomy", Work in Progress,
Internet-Draft, draft-ietf-detnet-dataplane-taxonomy-01, 6
July 2024, <https://datatracker.ietf.org/doc/html/draft-
ietf-detnet-dataplane-taxonomy-01>.
[I-D.ietf-detnet-scaling-requirements]
Liu, P., Li, Y., Eckert, T. T., Xiong, Q., Ryoo, J.,
zhushiyin, and X. Geng, "Requirements for Scaling
Deterministic Networks", Work in Progress, Internet-Draft,
draft-ietf-detnet-scaling-requirements-06, 22 May 2024,
<https://datatracker.ietf.org/doc/html/draft-ietf-detnet-
scaling-requirements-06>.
[I-D.p-6man-deterministic-eh]
Peng, S., "Deterministic Source Route Header", Work in
Progress, Internet-Draft, draft-p-6man-deterministic-eh-
00, 20 June 2024, <https://datatracker.ietf.org/doc/html/
draft-p-6man-deterministic-eh-00>.
[I-D.pb-6man-deterministic-crh]
Peng, S. and R. Bonica, "Deterministic Routing Header",
Work in Progress, Internet-Draft, draft-pb-6man-
deterministic-crh-00, 29 February 2024,
<https://datatracker.ietf.org/doc/html/draft-pb-6man-
deterministic-crh-00>.
[I-D.peng-6man-deadline-option]
Peng, S., Tan, B., and P. Liu, "Deadline Option", Work in
Progress, Internet-Draft, draft-peng-6man-deadline-option-
01, 11 July 2022, <https://datatracker.ietf.org/doc/html/
draft-peng-6man-deadline-option-01>.
[I-D.peng-6man-delay-options]
Peng, S., "Delay Options", Work in Progress, Internet-
Draft, draft-peng-6man-delay-options-00, 18 January 2024,
<https://datatracker.ietf.org/doc/html/draft-peng-6man-
delay-options-00>.
[I-D.peng-detnet-policing-jitter-control]
Peng, S., Liu, P., and K. Basu, "Policing Caused Jitter
Control Mechanism", Work in Progress, Internet-Draft,
Peng, et al. Expires 9 February 2025 [Page 55]
Internet-Draft Deadline Queueing Mechanism August 2024
draft-peng-detnet-policing-jitter-control-00, 18 January
2024, <https://datatracker.ietf.org/doc/html/draft-peng-
detnet-policing-jitter-control-00>.
[I-D.peng-lsr-deterministic-traffic-engineering]
Peng, S., "IGP Extensions for Deterministic Traffic
Engineering", Work in Progress, Internet-Draft, draft-
peng-lsr-deterministic-traffic-engineering-02, 24 June
2024, <https://datatracker.ietf.org/doc/html/draft-peng-
lsr-deterministic-traffic-engineering-02>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC2212] Shenker, S., Partridge, C., and R. Guerin, "Specification
of Guaranteed Quality of Service", RFC 2212,
DOI 10.17487/RFC2212, September 1997,
<https://www.rfc-editor.org/info/rfc2212>.
[RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black,
"Definition of the Differentiated Services Field (DS
Field) in the IPv4 and IPv6 Headers", RFC 2474,
DOI 10.17487/RFC2474, December 1998,
<https://www.rfc-editor.org/info/rfc2474>.
[RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z.,
and W. Weiss, "An Architecture for Differentiated
Services", RFC 2475, DOI 10.17487/RFC2475, December 1998,
<https://www.rfc-editor.org/info/rfc2475>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/info/rfc8174>.
[RFC8655] Finn, N., Thubert, P., Varga, B., and J. Farkas,
"Deterministic Networking Architecture", RFC 8655,
DOI 10.17487/RFC8655, October 2019,
<https://www.rfc-editor.org/info/rfc8655>.
[RFC8938] Varga, B., Ed., Farkas, J., Berger, L., Malis, A., and S.
Bryant, "Deterministic Networking (DetNet) Data Plane
Framework", RFC 8938, DOI 10.17487/RFC8938, November 2020,
<https://www.rfc-editor.org/info/rfc8938>.
Peng, et al. Expires 9 February 2025 [Page 56]
Internet-Draft Deadline Queueing Mechanism August 2024
[RFC9016] Varga, B., Farkas, J., Cummings, R., Jiang, Y., and D.
Fedyk, "Flow and Service Information Model for
Deterministic Networking (DetNet)", RFC 9016,
DOI 10.17487/RFC9016, March 2021,
<https://www.rfc-editor.org/info/rfc9016>.
[RFC9055] Grossman, E., Ed., Mizrahi, T., and A. Hacker,
"Deterministic Networking (DetNet) Security
Considerations", RFC 9055, DOI 10.17487/RFC9055, June
2021, <https://www.rfc-editor.org/info/rfc9055>.
[RFC9320] Finn, N., Le Boudec, J.-Y., Mohammadpour, E., Zhang, J.,
and B. Varga, "Deterministic Networking (DetNet) Bounded
Latency", RFC 9320, DOI 10.17487/RFC9320, November 2022,
<https://www.rfc-editor.org/info/rfc9320>.
22.2. Informative References
[EDF-algorithm]
"A framework for achieving inter-application isolation in
multiprogrammed, hard real-time environments", 1996,
<https://ieeexplore.ieee.org/document/896011>.
[EF-FIFO] "Fundamental Trade-Offs in Aggregate Packet Scheduling",
2001, <https://ieeexplore.ieee.org/document/992892>.
[IR-Theory]
"A Theory of Traffic Regulators for Deterministic Networks
with Application to Interleaved Regulators", 2018,
<https://ieeexplore.ieee.org/document/8519761>.
[Jitter-EDF]
"Delay Jitter Control for Real-Time Communication in a
Packet Switching Network", 1991,
<https://citeseerx.ist.psu.edu/document?repid=rep1&type=pd
f&doi=a2018cc8993c3705e851480a1b75addc7ce6bc9b>.
[Net-Calculus]
"Network Calculus: A Theory of Deterministic Queuing
Systems for the Internet", 2001,
<https://leboudec.github.io/netcal/latex/netCalBook.pdf>.
[P802.1DC] "Quality of Service Provision by Network Systems", 2023,
<https://1.ieee802.org/tsn/802-1dc/>.
[PIFO] "Programmable Packet Scheduling at Line Rate", 2016,
<https://dl.acm.org/doi/pdf/10.1145/2934872.2934899>.
Peng, et al. Expires 9 February 2025 [Page 57]
Internet-Draft Deadline Queueing Mechanism August 2024
[RPQ] "Exact Admission Control for Networks with a Bounded Delay
Service", 1996,
<https://ieeexplore.ieee.org/document/556345/>.
[SP-LATENCY]
"Guaranteed Latency with SP", 2020,
<https://ieeexplore.ieee.org/document/9249224>.
Appendix A. Proof of Schedulability Condition for RPQ
Figure 19 below gives the proof of schedulability condition for RPQ.
^
| Planned Residence Time
| | |
|CT_t+CTI+2*RTI -> ===== | |
| CT_t+CTI+RTI -> =====| |
CT_t + - - - - - - - - - - - - >+====+ -\
+CTI | | | |
| | | |
| | | > Phycical queue-x
D_p + - - - - - - - - - - - - >| | |
| | | |
CT_t + - - - - - - - - - - - - >+====+ -/
| | |===== <- CT_t-RTI
| | | ===== <- CT_t-2*RTI
| | |
|
| RTI| ... ... | RTI| RTI| RTI| RTI| RTI| RTI|
---+----+-----------+----+----+----+----+----+----+--------->
0 ^ ^ ^ ^ time
| | | |
t-tau' t-ofst t t+tau
(busy period begin) (arrival) (departure)
Figure 19: RPQ Based Scheduling
Suppose that the observed packet, with planned residence time D_p,
arrives at the scheduler at time t and leaves the scheduler at time
t+tau. It will be inserted to physical queue-x with count-down time
CT_t at the current timer interval RTI with starting time t-ofst and
end time t-ofst+RTI. According to the above packet queueing rules,
we have CT_t <= D_p < CT_t+CTI. Also suppose that t-tau' is the
beginning of the busy period closest to t. Then, we can get the
amount of packets within time interval [t-tau', t+tau] that must be
scheduled before the observed packet. In detailed:
Peng, et al. Expires 9 February 2025 [Page 58]
Internet-Draft Deadline Queueing Mechanism August 2024
* For all flow i with planned residence time D_i meeting CT_t <= D_i
< CT_t+CTI, the workload is sum{A'_i[t-tau', t]}.
Explanation: since the packets with planned residence time D_i
in the range [CT_t, CT_t+CTI) arrived at time t will be sent
before the observed packet, the packets with the same D_i
before time t will become more urgent at time t, and must also
be sent before the observed packet.
* For all flow i with planned residence time D_i meeting D_i >=
CT_t+CTI, the workload is sum{A'_i[t-tau', t-ofst-(D_i-CT_t-
CTI)]}.
Explanation: although the packets with planned residence time
D_i larger than CT_t+CTI arrived at time t will be sent after
the observed packet, but the packets with the same D_i before
time t, especially before time t-ofst-(D_i-CT_t-CTI), will
become more urgent at time t, and must be sent before the
observed packet.
* For all flow i with planned residence time D_i meeting D_i < CT_t,
the workload is sum{A'_i[t-tau', t+(CT_t-D_i)]}.
Explanation: the packets with planned residence time D_i less
than CT_t at time t will certainly be sent before the observed
packet, at a future time t+(CT_t-D_i) the packets with the same
D_i will still be urgent than the observed packet (even the
observed packet also become urgent), and must be sent before
the observed packet.
* Then deduct the traffic that has been sent during the busy period,
i.e., C*(tau+tau').
Let tau as D_p, and remember that CT_t <= D_p, the above workload is
less than
sum{A'_i(tau'+CT_t+CTI-D_i) for all D_i >= CT_t} +
sum{A'_i(tau'+CT_t-D_i) for all D_i < CT_t} - C*(tau'+D_p)
It is further less than
sum{A'_i(tau'+D_p+CTI-D_i) for all D_i >= D_2} + A'_1(tau'+D_p-
D_1) - C*(tau'+D_p)
Then, denote x as tau'+D_p, we have
sum{A'_i(x+CTI-D_i) for all D_i >= D_2} + A'_1(x-D_1) - C*(x)
Peng, et al. Expires 9 February 2025 [Page 59]
Internet-Draft Deadline Queueing Mechanism August 2024
In the case that d_i contains only one D_i, we have A_i = A'_i, d_i =
D_i, so the above workload is
sum{A_i(x+CTI-d_i) for all d_i >= d_2} + A_1(x-d_1) - C*(x)
Let the above workload be less than zero, then we get Equation-2.
In the case that d_i contains multiple D_i, e.g., d_1 is the minimum
delay level with 10us, D_1 ~ D_10 is 10 ~ 19us respectively, d_2 is
20us, D_11 ~ D_20 iS 20 ~29us respectively, etc. Let D_1 ~ D_10
consume the resources of d_1, and D_11 ~ D_20 consume the resources
of d_2, etc. Then, the above workload is less than
sum{A'_i(x+CTI-d_i) for all D_i belongs to d_i} - C*(x)
That is sum{A_i(x+CTI-d_i) for all d_i} - C*(x), and let it less than
zero, then we get Equation-3.
Appendix B. Proof of Schedulability Condition for Alternate QAR of RPQ
In the case that d_i contains only one D_i, the schedulability
condition is Equation-1. This is because, in the workload, for all
D_i meeting D_i >= CT_t+CTI, their contributed workload is changed to
sum{A'_i[t-tau', t-ofst-(D_i-CT_t)]} based on the analysis of
Equation-2, that is, the amount of workload A'_i(CTI) (that is placed
in queue-x) is excluded.
In the case that d_i contains multiple D_i, the schedulability
condition is still Equation-3. This is because multiple D_i may
belong to the same delay level as D_p. Assuming that within time
zone [t-ofst, t-ofst+I] the list of all arrived D_i in the same
parent queue-x with [CT_t, CT_t+CTI) as the observed packet (with
D_p) is:
* D_a1~D_am, where D_a1 is closer to CT_t+CTI, they are larger than
D_p (but smaller than CT_t+CTI) and belongs to a larger delay
level than d_p (corresponding delay level of D_p).
* D_b1~D_bm, they are larger than D_p and belongs to the same delay
level as d_p.
* D_p.
* D_c1~D_cm, they are smaller than D_p, and may belongs to the same
delay level as d_p or a lower delay level than d_p.
Peng, et al. Expires 9 February 2025 [Page 60]
Internet-Draft Deadline Queueing Mechanism August 2024
So that both D_b1~D_bm and D_c1~D_cm should be scheduled before the
observed packet. This is also true for these set of packets that
have arrived in history.
Strictly, for D_a1, the contributed workload is sum{A'_i[t-tau',
t-ofst+I-CTI]}, that is, only before time t-ofst+I-CTI the arrived
packets of D_a1 will be placed in a more urgent queue-y with [CT_t,
CT_t + CTI) than queue-x (at this history time its CT is [CT_t+CTI,
CT_t + 2AT)) and should be shceduled before the observed packet.
Similarly, for D_a2, the contributed workload is sum{A'_i[t-tau',
t-ofst+I-CTI+I]}, for D_am, the contributed workload is
sum{A'_i[t-tau', t-ofst+I-CTI+(m-1)*I]}.
Note that queue-x also contains packets with D_i (e.g., D_a0, larger
than D_a1) that have arrived in history. For D_a0, the contributed
workload is sum{A'_i[t-tau', t-ofst+I-CTI-(D_a0-D_a1)]}.
However, the number of m is not fixed. For safety, we can
appropriately overestimate workload time zone of D_a1~D_am to time
instant t and regard that they need to be scheduled before the
observed packet. Based on this, we can get the Equation-3.
Authors' Addresses
Shaofu Peng
ZTE Corporation
China
Email: peng.shaofu@zte.com.cn
Zongpeng Du
China Mobile
China
Email: duzongpeng@foxmail.com
Kashinath Basu
Oxford Brookes University
United Kingdom
Email: kbasu@brookes.ac.uk
Zuopin Cheng
New H3C Technologies
China
Email: czp@h3c.com
Peng, et al. Expires 9 February 2025 [Page 61]
Internet-Draft Deadline Queueing Mechanism August 2024
Dong Yang
Beijing Jiaotong University
China
Email: dyang@bjtu.edu.cn
Chang Liu
China Unicom
China
Email: liuc131@chinaunicom.cn
Peng, et al. Expires 9 February 2025 [Page 62]