TSVWG Y. Li
Internet-Draft X. Zhou
Intended status: Informational Huawei
Expires: November 25, 2019 May 24, 2019
LOOPS (Localized Optimizations of Path Segments) Problem Statement and
Opportunities
draft-li-tsvwg-loops-problem-opportunities-02
Abstract
In various network deployments, end to end paths are partitioned into
multiple segments. In some cloud based WAN connections, multiple
overlay tunnels in series are used to achieve better path selection
and lower latency. In satellite communication, the end to end path
is split into two terrestrial segments and a satellite segment.
Packet losses can be caused both by random events or congestion in
various deployments.
Traditional end-to-end transport layers respond to packet loss slowly
especially in long-haul networks: They either wait for some signal
from the receiver to indicate a loss and then retransmit from the
sender or rely on sender's timeout which is often quite long. Non-
congestion caused packet loss may make the TCP sender over-reduce the
sending rate unnecessarily. With end-to-end encryption moving under
the transport (QUIC), traditional PEP (performance enhancing proxy)
techniques such as TCP splitting are no longer applicable.
LOOPS (Local Optimizations on Path Segments) aims to provide non end-
to-end, locally based in-network recovery to achieve better data
delivery by making packet loss recovery faster and by avoiding the
senders over-reducing their sending rate. In an overlay network
scenario, LOOPS can be performed over the existing, or purposely
created, overlay tunnel based path segments.
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
Li & Zhou Expires November 25, 2019 [Page 1]
Internet-Draft LOOPS Problem & opportunities May 2019
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 November 25, 2019.
Copyright Notice
Copyright (c) 2019 IETF Trust and the persons identified as the
document authors. All rights reserved.
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 Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1. Terminology . . . . . . . . . . . . . . . . . . . . . . . 4
2. Cloud-Internet Overlay Network . . . . . . . . . . . . . . . 5
2.1. Tail Loss or Loss in Short Flows . . . . . . . . . . . . 7
2.2. Packet Loss in Real Time Media Streams . . . . . . . . . 8
2.3. Packet Loss and Congestion Control in Bulk Data Transfer 8
2.4. Multipathing . . . . . . . . . . . . . . . . . . . . . . 9
3. Satellite Communication . . . . . . . . . . . . . . . . . . . 9
4. Features and Impacts to be Considered for LOOPS . . . . . . . 11
4.1. Local Recovery and End-to-end Retransmission . . . . . . 12
4.1.1. OE to OE Measurement, Recovery and Multipathing . . . 13
4.2. Congestion Control Interaction . . . . . . . . . . . . . 14
4.3. Overlay Protocol Extensions . . . . . . . . . . . . . . . 16
4.4. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 16
5. Security Considerations . . . . . . . . . . . . . . . . . . . 17
6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 17
7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 17
8. Informative References . . . . . . . . . . . . . . . . . . . 17
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 19
1. Introduction
Overlay tunnels are widely deployed for various networks, including
long haul WAN interconnection, enterprise wireless access networks,
etc. The end to end connection is partitioned into multiple path
segments using overlay tunnels. This serves a number of purposes,
Li & Zhou Expires November 25, 2019 [Page 2]
Internet-Draft LOOPS Problem & opportunities May 2019
for instance, selecting a better path over the WAN or delivering the
packets over heterogeneous network, such as enterprise access and
core networks.
A reliable transport layer normally employs some end-to-end
retransmission mechanisms which also address congestion control
[RFC0793] [RFC5681]. The sender either waits for the receiver to
send some signals on a packet loss or sets some form of timeout for
retransmission. For unreliable transport layer protocols such as RTP
[RFC3550], optional and limited usage of end-to-end retransmission is
employed to recover from packet loss [RFC4585] [RFC4588].
End-to-end retransmission to recover lost packets is slow especially
when the network is long haul. When a path is partitioned into
multiple path segments that are realized as overlay tunnels, LOOPS
(Local Optimizations on Path Segments) tries to provide local segment
based in-network recovery to achieve better data delivery by making
packet loss recovery faster and by avoiding the senders over-reducing
their sending rate. In an overlay network scenario, LOOPS can be
performed over the existing, or purposely created, overlay tunnel
based path segments.
Some link types (satellite, microwave) may exhibit unusually high
loss rate in special conditions (e.g., fades due to heavy rain). The
traditional TCP sender interprets loss as congestion and over-reduces
the sending rate, degrading the throughput. LOOPS is also applicable
to such scenarios to improve throughput.
Section 2 presents some of the issues and opportunities found in
Cloud-Internet overlay networks that require higher performance and
more reliable packet transmission in best effort networks. Section 3
discusses applications of LOOPS in satellite communication.
Section 4 describes the corresponding solution features and the their
impact on existing network technologies.
Li & Zhou Expires November 25, 2019 [Page 3]
Internet-Draft LOOPS Problem & opportunities May 2019
ON=overlay node
UN=underlay node
+---------+ +---------+
| App | <---------------- end-to-end ---------------> | App |
+---------+ +---------+
|Transport| <---------------- end-to-end ---------------> |Transport|
+---------+ +---------+
| | | |
| | +--+ path +--+ path segment2 +--+ | |
| | | |<-seg1->| |<--------------> | | | |
| Network | +--+ |ON| +--+ |ON| +--+ +----+ |ON| | Network |
| |--|UN|--| |--|UN|--| |--|UN|---| UN |--| |--| |
+---------+ +--+ +--+ +--+ +--+ +--+ +----+ +--+ +---------+
End Host End Host
<--------------------------------->
LOOPS domain: path segment enables
optimizations for better local transport
Figure 1: LOOPS in Overlay Network Usage Scenario
1.1. Terminology
LOOPS: Local Optimizations on Path Segments. LOOPS includes the
local in-network (i.e. non end-to-end) recovery function, for
instance, loss detection and measurements.
LOOPS Node: Node supporting LOOPS functions.
Overlay Node (ON): Node having overlay functions (like overlay
protocol encapsulation/decapsulation, header modification, TLV
inspection) and LOOPS functions in LOOPS overlay network usage
scenario. Both OR and OE are Overlay Nodes.
Overlay Tunnel: A tunnel with designated ingress and egress nodes
using some network overlay protocol as encapsulation, optionally
with a specific traffic type.
Overlay Path: A channel within the overlay tunnel, where the traffic
transmitted on the channel needs to pass through zero or more
designated intermediate overlay nodes. There may be more than one
overlay path within an overlay tunnel when the different sets of
designated intermediate overlay nodes are specified. An overlay
path may contain multiple path segments. When an overlay tunnel
contains only one overlay path without any intermediate overlay
node specified, overlay path and overlay tunnel are used
interchangeably.
Li & Zhou Expires November 25, 2019 [Page 4]
Internet-Draft LOOPS Problem & opportunities May 2019
Overlay Edge (OE): Edge node of an overlay tunnel.
Overlay Relay (OR): Intermediate overlay node on an overlay path.
An overlay path need not contain any OR.
Path segment: Part of an overlay path between two neighbor overlay
nodes. It is used interchangeably with overlay segment in this
document when the context wants to emphasize on its overlay
encapsulated nature. An overlay path may contain multiple path
segments. When an overlay path contains only one path segment,
i.e. the segment is between two OEs, the path segment is
equivalent to the overlay path. It is also called segment for
simplicity in this document.
Overlay segment: Refers to path segment.
Underlay Node (UN): Nodes not participating in the overlay network
function.
2. Cloud-Internet Overlay Network
The Internet is a huge network of networks. The interconnections of
end devices using this global network are normally provided by ISPs
(Internet Service Provider). This network created by the composition
of the ISP networks is considered as the traditional Internet. CSPs
(Cloud Service Providers) are connecting their data centers using the
Internet or via self-constructed networks/links. This expands the
Internet's infrastructure and, together with the original ISP's
infrastructure, forms the Internet underlay.
NFV (network function virtualization) further makes it easier to
dynamically provision a new virtual node as a work load in a cloud
for CPU/storage intensive functions. With the aid of various
mechanisms such as kernel bypassing and Virtual IO, forwarding based
on virtual nodes is becoming more and more effective. The
interconnections among the purposely positioned virtual nodes and/or
the existing nodes with virtualization functions potentially form an
overlay of Internet. It is called the Cloud-Internet Overlay Network
(CION) in this document.
CION makes use of overlay technologies to direct the traffic going
through the specific overlay path regardless of the underlying
physical topology, in order to achieve better service delivery. It
purposely creates or selects overlay nodes (ON) from providers. By
continuously measuring the delay of path segments and use them as
metrics for path selection, when the number of overlay nodes is
sufficiently large, there is a high chance that a better path could
be found [DOI_10.1109_ICDCS.2016.49] [DOI_10.1145_3038912.3052560].
Li & Zhou Expires November 25, 2019 [Page 5]
Internet-Draft LOOPS Problem & opportunities May 2019
[DOI_10.1145_3038912.3052560] further shows all cloud providers
experience random loss episodes and random loss accounts for more
than 35% of total loss.
Figure 2 shows an example of an overlay path over large geographic
distances. The path between two OEs (Overlay Edges) is an overlay
path. OEs are ON1 & ON4 in Figure 2. Part of the path between ONs
is a path segment. Figure 2 shows the overlay path with 3 segments,
i.e. ON1-ON2-ON3-ON4. ON is usually a virtual node, though it does
not have to be. Overlay path transmits packets in some form of
network overlay protocol encapsulation. ON has the computing and
memory resources that can be used for some functions like packet loss
detection, network measurement and feedback, packet recovery.
_____________
/ domain 1 \
/ \
___/ -------------\
/ \
PoP1 ->--ON1 \
| | ON4------>-- PoP2
| | ON2 ___|__/
\__|_ |->| _____ / |
| \|__|__ / \ / |
| | | \____/ \__/ |
\|/ | | _____ |
| | | ___/ \ |
| | \|/ / \_____ |
| | | / domain 2 \ /|\
| | | | ON3 | |
| | | \ |->| | |
| | | \_____|__|_______/ |
| /|\ | | \|/ |
| | | | | |
| | | /|\ | |
+--------------------------------------------------+
| | | | | | | Internet |
| o--o o---o->---o o---o->--o--o underlay |
+--------------------------------------------------+
Figure 2: Cloud-Internet Overlay Network (CION)
We tested based on 37 overlay nodes from multiple cloud providers
globally. Each pair of the overlay nodes are used as sender and
receiver. When the traffic is not intentionally directed to go
through any intermediate virtual nodes, we call the path that the
traffic takes the _default path_ in the test. When any of the
virtual nodes is intentionally used as an intermediate node to
Li & Zhou Expires November 25, 2019 [Page 6]
Internet-Draft LOOPS Problem & opportunities May 2019
forward the traffic, the path that the traffic takes is an _overlay
path_ in the test. The preliminary experiments showed that the delay
of an overlay path is shorter than that of the default path in 69% of
cases at 99% percentile and improvement is 17.5% at 99% percentile
when we probe Ping packets every second for a week.
Lower delay does not necessarily mean higher throughput. Different
path segments may have different packet loss rates. Loss rate is
another major factor impacting TCP throughput. From some customer
requirements, we set the target loss rate to be less than 1% at 99%
percentile and 99.9% percentile, respectively. The loss was measured
between any two overlay nodes, i.e. any potential path segment. Two
thousand Ping packets were sent every 20 seconds between two overlay
nodes for 55 hours. This preliminary experiment showed that the
packet loss rate satisfaction are 44.27% and 29.51% at the 99% and
99.9% percentiles respectively.
Hence packet loss in an overlay segment is a key issue to be solved
in CION. In long-haul networks, the end-to-end retransmission of
lost packet can result in an extra round trip time. Such extra time
is not acceptable in some cases. As CION naturally consists of
multiple overlay segments, LOOPS leverages this to perform local
optimizations on a single hop between two overlay nodes. ("Local"
here is a concept relative to end-to-end, it does not mean such
optimization is limited to LAN networks.)
The following subsections present different scenarios using multiple
segment based overlay paths with a common need of local in-network
loss recovery in best effort networks.
2.1. Tail Loss or Loss in Short Flows
When the lost segments are at the end of a transaction, TCP's fast
retransmit algorithm does not work as there are no ACKs to trigger
it. When a sender does not receive an ACK for a given segment within
a certain amount of time called retransmission timeout (RTO), it re-
sends the segment [RFC6298]. RTO can be as long as several seconds.
Hence the recovery of lost segments triggered by RTO is lengthy.
[I-D.dukkipati-tcpm-tcp-loss-probe] indicates that large RTOs make a
significant contribution to the long tail on the latency statistics
of short flows like web pages.
The short flow often completes in one or two RTTs. Even when the
loss is not a tail loss, it can possibly add another RTT because of
end-to-end retransmission (not enough packets are in flight to
trigger fast retransmit). In long haul networks, it can result in
extra time of tens or even hundreds of milliseconds.
Li & Zhou Expires November 25, 2019 [Page 7]
Internet-Draft LOOPS Problem & opportunities May 2019
An overlay segment transmits the aggregated flows from ON to ON. As
short flows are aggregated, the probability of tail loss over this
specific overlay segment decreases compared to an individual flow.
The overlay segment is much shorter than the end-to-end path in a
Cloud- Internet overlay network, hence loss recovery over an overlay
segment is faster.
2.2. Packet Loss in Real Time Media Streams
The Real-time transport protocol (RTP) is widely used in interactive
audio and video. Packet loss degrades the quality of the received
media. When the latency tolerance of the application is sufficiently
large, the RTP sender may use RTCP NACK feedback from the receiver
[RFC4585] to trigger the retransmission of the lost packets before
the playout time is reached at the receiver.
In a Cloud-Internet overlay network, the end-to-end path can be
hundreds of milliseconds. End-to-end feedback based retransmission
may be not be very useful when applications can not tolerate one more
RTT of this length. Loss recovery over an overlay segment can then
be used for the scenarios where RTCP NACK triggered retransmission is
not appropriate.
2.3. Packet Loss and Congestion Control in Bulk Data Transfer
TCP congestion control algorithms such as Reno and CUBIC basically
interpret packet loss as congestion experienced somewhere in the
path. When a loss is detected, the congestion window will be
decreased at the sender to make the sending slower. It has been
observed that packet loss is not an accurate way to detect congestion
in the current Internet [I-D.cardwell-iccrg-bbr-congestion-control].
In long-haul links, when the loss is caused by non-persistent burst
which is extremely short and pretty random, the sender's reaction of
reducing sending rate is not able to respond in time to the
instantaneous path situation or to mitigate such bursts. On the
contrary, reducing window size at the sender unnecessarily or too
aggressively harms the throughput for application's long lasting
traffic like bulk data transfer.
The overlay nodes are distributed over the path with computing
capability, they are in a better position than the end hosts to
deduce the underlying links' instantaneous situation from measuring
the delay, loss or other metrics over the segment. Shorter round
trip time over a path segment will benefit more accurate and
immediate measurements for the maximum recent bandwidth available,
the minimum recent latency, or trend of change. ONs can further
decide if the sending rate reduction at the sender is necessary when
a loss happened. Section 4.2 talks more details on this.
Li & Zhou Expires November 25, 2019 [Page 8]
Internet-Draft LOOPS Problem & opportunities May 2019
2.4. Multipathing
As an overlay path may suffer from an impairment of the underlying
network, two or more overlay paths between the same set of ingress
and egress overlay nodes can be combined for reliability purpose.
During a transient time when a network impairment is detected,
sending replicating traffic over two paths can improve reliability.
When two or more disjoint overlay paths are available as shown in
Figure 3 from ON1 to ON2, different sets of traffic may use different
overlay paths. For instance, one path is for low latency and the
other is for higher bandwidth, or they can be simply used as load
balancing for better bandwidth utilization.
Two disjoint paths can usually be found by measuring to figure out
the segments with very low mathematical correlation in latency
change. When the number of overlay nodes is large, it is easy to
find disjoint or partially disjoint segments.
Different overlay paths may have varying characteristics. The
overlay tunnel should allow the overlay path to handle the packet
loss depending on its own path measurements.
ON-A
+----------o------------------+
| |
| |
A -----o ON1 ON2o----- B
| |
+-----------------------o-----+
ON-B
Figure 3: Multiple Overlay Paths
3. Satellite Communication
Traditionally, satellite communications deploy PEP (performance
enhancing proxy) nodes around the satellite link to enhance end-to-
end performance. TCP splitting is a common approach employed by such
PEPs, where the TCP connection is split into three: the segment
before the satellite hop, the satellite section (uplink, downlink),
and the segment behind the satellite hop. This requires heavy
interactions with the end-to-end transport protocols, usually without
the explicit consent of the end hosts. Unfortunately, this is
indistinguishable from a man-in-the-middle attack on TCP. With end-
to-end encryption moving under the transport (QUIC), this approach is
no longer useful.
Li & Zhou Expires November 25, 2019 [Page 9]
Internet-Draft LOOPS Problem & opportunities May 2019
Geosynchronous Earth Orbit (GEO) satellites have a one-way delay (up
to the satellite and back) on the order of 250 milliseconds. This
does not include queueing, coding and other delays in the satellite
ground equipment. The Round Trip Time for a TCP or QUIC connection
going over a satellite hop in both directions, in the best case, will
be on the order of 600 milliseconds. And, it may be considerably
longer. RTTs on this order of magnitude have significant performance
implications.
Packet loss recovery is an area where splitting the TCP connection
into different parts helps. Packets lost on the terrestrial links
can be recovered at terrestrial latencies. Packet loss on the
satellite link can be recovered more quickly by an optimized for
satellite protocol between the PEPs and/or link layer FEC than they
could be end to end. Again, encryption makes TCP splitting no longer
applicable. Enhanced error recovery at the satellite link layer
helps for the loss on the satellite link but doesn't help for the
terrestrial links. Even when the terrestrial segments are short, any
loss must be recovered across the satellite link delay. And, there
are cases when a satellite ground station connects to the general
Internet with a potentially larger terrestrial segment (e.g., to a
correspondent host in another country). Faster recovery over such
long terrestrial segments is desirable.
Another aspect of recovery is that terrestrial loss is highly likely
to be congestion related but satellite loss is more likely to be
transmission errors due to link conditions. A transport endpoint
slowing down because of mis-interpreting these errors as congestion
losses unnecessarily reduces performance. But, at the end points,
the difference between the two is not easily distinguished. To
elaborate more on the loss recovery for satellite communications,
while the error rate on the satellite paths is generally very low
most of the time, it might get higher during special link conditions
(e.g. fades due to heavy rain). The satellite hop itself does know
which losses are due to link conditions as opposed to congestion, but
it has no mechanism to signal this difference to the end hosts.
We will need the protocol under QUIC to try to minimize non-
congestion packet drop. Specific link layers may have techniques
such as satellite FEC to recover. Where the capabilities of that may
be exceeded (e.g., rain fade), we can look at LOOPS-like approaches.
There are two high level classes of solutions for making encrypted
transport traffic like QUIC work well over satellite:
o Hooks in the protocol which can adapt to large BDPs where both the
bandwidth and the latency are large. This would require end to
end enhancement.
Li & Zhou Expires November 25, 2019 [Page 10]
Internet-Draft LOOPS Problem & opportunities May 2019
o Capabilities (such as LOOPS) under the protocol to improve
performance over specific segments of the path. In particular,
separating the terrestrial from the satellite losses. Fixing the
terrestrial loss quickly and keeping throughput high over
satellite segment by not causing the end-hosts to over-reduce
their sending window in case of non-congestion loss.
This document focuses on the latter.
4. Features and Impacts to be Considered for LOOPS
LOOPS (Localized Optimizations of Path Segments) aims to leverage the
virtual nodes in a selected path to improve the transport performance
"locally" instead of end-to-end as those nodes have partitioned the
path to multiple segments. With the technologies like NFV (Network
function virtualization) and virtual IO, it is easier to add
functions to virtual nodes and even the forwarding on those virtual
nodes is getting more efficient. Some overlay protocols such as
VXLAN [RFC7348], GENEVE [I-D.ietf-nvo3-geneve], LISP [RFC6830] or
CAPWAP [RFC5415] are assumed to be employed in the network. In
overlay network usage scenario, LOOPS can extend a specific overlay
protocol header to perform local measurement and local recovery
functions, like the example shown in Figure 4.
+------------+------------+-----------------+---------+---------+
|Outer IP hdr|Overlay hdr |LOOPS information|Inner hdr|payload |
+------------+------------+-----------------+---------+---------+
Figure 4: LOOPS Extension Header Example
LOOPS uses packet number space independent from that of the transport
layer. Acknowledgment should be generated from ON receiver to ON
sender for packet loss detection and local measurement. To reduce
overhead, negative ACK over each path segment is a good choice here.
A Timestamp echo mechanism, analogous to TCP's Timestamp option,
should be employed in band in LOOPS extension to measure the local
RTT and variation for an overlay segment. Local in-network recovery
is performed. The measurement over segment is expected to give a
hint on whether the lost packet of locally recovered one was caused
by congestion. Such a hint could be further feedback, using like by
ECN Congestion Experienced (CE) markings, to the end host sender. It
directs the end host sender if congestion window adjustment is
necessary. LOOPS normally works on the overlay segment which
aggregates the same type of traffic, for instance TCP traffic or
finer granularity like TCP throughput sensitive traffic. LOOPS does
not look into the inner packet. Elements to be considered in LOOPS
are discussed briefly here.
Li & Zhou Expires November 25, 2019 [Page 11]
Internet-Draft LOOPS Problem & opportunities May 2019
4.1. Local Recovery and End-to-end Retransmission
There are basically two ways to perform local recovery,
retransmission and FEC (forward error correction). They are possibly
used together in some cases. Such approaches between two overlay
nodes recover the lost packet in relatively shorter distance and thus
shorter latency. Therefore the local recovery is always faster
compared to end-to- end.
At the same time, most transport layer protocols have their own end-
to-end retransmission to recover the lost packet. It would be ideal
that end-to-end retransmission at the sender was not triggered if the
local recovery was successful.
End-to-end retransmission is normally triggered by a NACK as in RTCP
or multiple duplicate ACKs as in TCP.
When FEC is used for local recovery, it may come with a buffer to
make sure the recovered packets delivered are in order subsequently.
Therefore the receiver side is unlikely to see the out-of-order
packets and then send a NACK or multiple duplicate ACKs. The side
effect to unnecessarily trigger end-to-end retransmit is minimum.
When FEC is used, if redundancy and block size are determined, extra
latency required to recover lost packets is also bounded. Then RTT
variation caused by it is predictable. In some extreme case like a
large number of packet loss caused by persistent burst, FEC may not
be able to recover it. Then end-to-end retransmit will work as a
last resort. In summary, when FEC is used as local recovery, the
impact on end-to-end retransmission is limited.
When retransmission is used, more care is required.
For packet loss in RTP streaming, retransmission can recover those
packets which would not be retransmitted end-to-end otherwise due to
long RTT. It would be ideal if the retransmitted packet reaches the
receiver before it sends back information that the sender would
interpret as a NACK for the lost packet. Therefore when the
segment(s) being retransmitted is a small portion of the whole end to
end path, the retransmission will have a significant effect of
improving the quality at receiver. When the sender also re-transmits
the packet based on a NACK received, the receiver will receive the
duplicated retransmitted packets and should ignore the duplication.
For packet loss in TCP flows, TCP RENO and CUBIC use duplicate ACKs
as a loss signal to trigger the fast retransmit. There are different
ways to avoid the sender's end-to-end retransmission being triggered
prematurely:
Li & Zhou Expires November 25, 2019 [Page 12]
Internet-Draft LOOPS Problem & opportunities May 2019
o The egress overlay node can buffer the out-of-order packets for a
while, giving a limited time for a packet being retransmitted
somewhere in the overlay path to reach it. The retransmitted
packet and the buffered packets caused by it may increase the RTT
variation at the sender. When the retransmitted latency is a
small portion of RTT or the loss is rare, such RTT variation will
be smoothed without much impact. Another possible way is to make
the sender exclude such packets from the RTT measurement. The
locally recovered packets can be specially marked and this marking
is spin back to end host sender. Then RTT measurement should not
use that packet.
The buffer management is nontrivial in this case. It has to be
determined how many out-of-order packets can be buffered at the
egress overlay node before it gives up waiting for a successful
local retransmission. As the lost packet is not always recovered
successfully locally, the sender may invoke end-to-end fast
retransmit slower than it would be in classic TCP.
o If LOOPS network does not buffer the out-of-order packets caused
by packet loss, TCP sender can use a time based loss detection
like RACK [I-D.ietf-tcpm-rack] to prevent the TCP sender from
invoking fast retransmit too early. RACK uses the notion of time
to replace the conventional DUPACK threshold approach to detect
losses. RACK is required to be tuned to fit the local
retransmission better. If there are n similar segments over the
path, segment retransmission will at least add RTT/n to the
reordering window by average when the packet is lost only once
over the whole overlay path. This approach is more preferred than
one described in previous bullet. On the other hand, if time
based loss detection is not supported at the sender, end to end
retransmission will be invoked as usual. It wastes some
bandwidth.
4.1.1. OE to OE Measurement, Recovery and Multipathing
When local recovery is between two neighbor ONs, it is called per-hop
recovery. It can be between overlay relays or between overlay relay
and overlay edge. Another type of local recovery is called OE to OE
recovery which performs between overlay edge nodes. When the
segments of an overlay path have similar characteristics and/or only
OE has the expected processing capability, OE to OE based local
recovery can be used instead of per-hop recovery.
If there is more than one overlay path in an overlay tunnel,
multipathing splits and recombines the traffic. Measurements such as
round trip time and loss rate between OEs hav to be specific to each
path. The ingress OE can use the feedback measurement to determine
Li & Zhou Expires November 25, 2019 [Page 13]
Internet-Draft LOOPS Problem & opportunities May 2019
the FEC parameter settings for different path. FEC can also be
configured to work over the combined path. The egress OE must be
able to remove the replicated packet when overlay path is switched
during impairment.
OE to OE measurement can help each segment determine its proportion
in edge to edge delay. It is useful for ON to decide if it is
necessary to turn on the per-hop recovery or how to fine tune the
parameter settings. When the segment delay ratio is small, the
segment retransmission is more effective.
4.2. Congestion Control Interaction
When a TCP-like transport layer protocol is used, local recovery in
LOOPS has to interact with the upper layer transport congestion
control. Classic TCP adjusts the congestion window when a loss is
detected and fast retransmit is invoked.
The local recovery mechanism breaks the assumption of the necessary
and sufficient conditional relationship between detected packet loss
and congestion control trigger at the sender in classic TCP. The
loss that is locally recovered can be caused by a non-persistent
congestion such as a microburst or a random loss, both of which
ideally would not let the sender invoke the congestion control
mechanism. But then, it can also possibly caused by a real
persistent congestion which should let the sender invoke sending rate
reduction. In either case, the sender does not see the locally
recovered packet as a loss.
When the local recovery takes effect, we consider the following two
cases. Firstly, the classic TCP sender does not see the enough
number of duplicate ACKs to trigger fast retransmit. This could be
the result of in-order packet delivery including locally recovered
ones to the receiver as mentioned in last subsection. Classic TCP
sender in this case will not reduce congestion window as no loss is
detected. Secondly, if a time based loss detection such as RACK is
used, as long as the locally recovered packet's ACK reaches the
sender before the reordering window expires, the congestion window
will not be reduced.
Such behavior brings the desirable throughput improvement when the
recovered packet is lost due to non-persistent congestion. It solves
the throughput problem mentioned in Section 2.3 and Section 3.
However, it also brings the risk that the sender is not able to
detect the real persistent congestion in time and then overshoot.
Eventually a severe congestion that is not recoverable by a local
recovery mechanism may occur. In addition, it may be unfriendly to
Li & Zhou Expires November 25, 2019 [Page 14]
Internet-Draft LOOPS Problem & opportunities May 2019
other flows (possibly pushing them out) if those flows are running
over the same underlying bottleneck links.
There is a spectrum of approaches. On one end, each locally
recovered packet can be treated exactly as a loss in order to invoke
the congestion control at the sender to guarantee the fair sharing as
classic TCP by setting its CE (Congestion Experienced) bit. Explicit
Congestion Notification (ECN) can be used here as ECN marking was
required to be equivalent to a packet drop [RFC3168]. Congestion
control at the sender works as usual and no throughput improvement
could be achieved (although the benefit of faster recovery is still
there). On the other hand, ON can perform its congestion measurement
over the segment, for instance local RTT and its variation trend.
Then the lost packet can be determined if it was caused by congestion
or other factors. It will further decide if it is necessary to set
CE marking or even what ratio is set to make the sender adjust the
sending rate more correctly.
There are possible cases that the sender detects the loss even with
local recovery in function. For example, when the re-ordering window
in RACK is not optimally adapted, the sender may trigger the
congestion control at the same time of end-to-end retransmission. If
spurious retransmission detection based on DSACK [RFC3708] is used,
such end-to-end retransmission will be found out unnecessary when
locally recovered packets reaches the receiver successfully. Then
congestion control changes will be undone at the sender. This
results in similar pros and cons as described earlier. Pros are
preventing the unnecessary window reduction and improving the
throughput when the loss is caused by non-persistent congestion or
random loss. Cons are some mechanisms like ECN or its variants
should be used wisely to make sure the congestion control is invoked
in case of persistent congestion.
An approach where the losses on a path segment are not immediately
made known to the end-to-end congestion control can be combined with
a "circuit breaker" style congestion control on the path segment.
When the usage of path segment by the overlay flow starts to become
unfair, the path segment sends congestion signals up to the end-to-
end congestion control. This must be carefully tuned to avoid
unwanted oscillation.
In summary, local recovery can improve Flow Completion Time (FCT) by
eliminating tail loss in small flows. As it changes loss event to
out-of-order event in most cases to TCP sender, if TCP sender uses
loss based congestion control, there is some implication on the
throughput. We suggest ECN and spurious retransmission to be enabled
when local recovery is in use, it would give the desirable
throughput, i.e. when loss is caused by congestion, reduce congestion
Li & Zhou Expires November 25, 2019 [Page 15]
Internet-Draft LOOPS Problem & opportunities May 2019
window; otherwise keep sender's sending rate. We do not suggest to
use spurious retransmission alone together with local recovery as it
may cause the TCP sender falsely undo window reduction when
congestion occurs. If only ECN is enabled or neither ECN nor
spurious retransmission is enabled, the throughput with local
recovery in use is no much difference from that of the tradition TCP.
4.3. Overlay Protocol Extensions
The overlay usually has no control over how packets are routed in the
underlying network between two overlay nodes, but it can control, for
example, the sequence of overlay nodes a message traverses before
reaching its destination. LOOPS assumes the overlay protocol can
deliver the packets in such designated sequence. Most forms of
overlay networking use some sort of "encapsulation". The whole path
taken can be performed by stitching multiple short overlay paths,
like VXLAN [RFC7348], GENEVE [I-D.ietf-nvo3-geneve], or it can be a
single overlay path with a sequence of intermediate overlay nodes
specified, as in SRv6 [I-D.ietf-6man-segment-routing-header]. In
either way, LOOPS information is required to be embedded in those
protocols to support the data plane measurement and feedback.
Retransmission or FEC based loss recovery can be either per ON-hop
based or OE to OE based.
LOOPS alone has no setup requirement on control plane. Some overlay
protocol, e.g. CAPWAP [RFC5415], has session setup phase, we can use
it to exchange the information such as dynamic FEC parameters.
4.4. Summary
LOOPS is expected to extend the existing overlay protocols in data
plane. Path selection is assumed a feature provided by the overlay
protocols via SDN or other approaches and is not a part of LOOPS.
LOOPS is a set of functions to be implemented on ONs in a long haul
overlay network. LOOPS includes the following features.
1. Local recovery. Retransmission, FEC or hybrid can be used as
local recovery method. Such recovery mechanism is in-network.
It is performed by two network nodes with computing and memory
resources.
2. Local congestion measurement. Sender ON measures the local
segment RTT, loss and/or throughput to immediately get the
overlay segment status.
3. Signal to end to end congestion control. Strategy to set/not set
ECN CE marking or simply drop the packet to signal the end host
sender about the loss event to help adjust the sending rate.
Li & Zhou Expires November 25, 2019 [Page 16]
Internet-Draft LOOPS Problem & opportunities May 2019
5. Security Considerations
LOOPS does not look at the traffic payload, so encrypted payload does
not affect functionality of LOOPS. The use of LOOPS introduces some
issues which impact security. ON with LOOPS function represents a
point in the network where the traffic can be potentially
manipulated. Denial of service attack can be launched from an ON. A
rogue ON might be able to spoof packet as if it come from a
legitimate ON. It may also modify the ECN CE marking in packets to
influence the sender's rate. In order to protected from such
attacks, the overlay protocol itself should have some build-in
security protection which inherently be used by LOOPS. The operator
should use some authentication mechanism to make sure ONs are valid
and non-compromised.
6. IANA Considerations
No IANA action is required.
7. Acknowledgements
Thanks to etosat mailing list about the discussion about the SatCom
and LOOPS use case.
8. Informative References
[RFC0793] Postel, J., "Transmission Control Protocol", STD 7,
RFC 793, DOI 10.17487/RFC0793, September 1981,
<https://www.rfc-editor.org/info/rfc793>.
[RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
of Explicit Congestion Notification (ECN) to IP",
RFC 3168, DOI 10.17487/RFC3168, September 2001,
<https://www.rfc-editor.org/info/rfc3168>.
[RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V.
Jacobson, "RTP: A Transport Protocol for Real-Time
Applications", STD 64, RFC 3550, DOI 10.17487/RFC3550,
July 2003, <https://www.rfc-editor.org/info/rfc3550>.
[RFC3708] Blanton, E. and M. Allman, "Using TCP Duplicate Selective
Acknowledgement (DSACKs) and Stream Control Transmission
Protocol (SCTP) Duplicate Transmission Sequence Numbers
(TSNs) to Detect Spurious Retransmissions", RFC 3708,
DOI 10.17487/RFC3708, February 2004,
<https://www.rfc-editor.org/info/rfc3708>.
Li & Zhou Expires November 25, 2019 [Page 17]
Internet-Draft LOOPS Problem & opportunities May 2019
[RFC4585] Ott, J., Wenger, S., Sato, N., Burmeister, C., and J. Rey,
"Extended RTP Profile for Real-time Transport Control
Protocol (RTCP)-Based Feedback (RTP/AVPF)", RFC 4585,
DOI 10.17487/RFC4585, July 2006,
<https://www.rfc-editor.org/info/rfc4585>.
[RFC4588] Rey, J., Leon, D., Miyazaki, A., Varsa, V., and R.
Hakenberg, "RTP Retransmission Payload Format", RFC 4588,
DOI 10.17487/RFC4588, July 2006,
<https://www.rfc-editor.org/info/rfc4588>.
[RFC5415] Calhoun, P., Ed., Montemurro, M., Ed., and D. Stanley,
Ed., "Control And Provisioning of Wireless Access Points
(CAPWAP) Protocol Specification", RFC 5415,
DOI 10.17487/RFC5415, March 2009,
<https://www.rfc-editor.org/info/rfc5415>.
[RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion
Control", RFC 5681, DOI 10.17487/RFC5681, September 2009,
<https://www.rfc-editor.org/info/rfc5681>.
[RFC6298] Paxson, V., Allman, M., Chu, J., and M. Sargent,
"Computing TCP's Retransmission Timer", RFC 6298,
DOI 10.17487/RFC6298, June 2011,
<https://www.rfc-editor.org/info/rfc6298>.
[RFC6830] Farinacci, D., Fuller, V., Meyer, D., and D. Lewis, "The
Locator/ID Separation Protocol (LISP)", RFC 6830,
DOI 10.17487/RFC6830, January 2013,
<https://www.rfc-editor.org/info/rfc6830>.
[RFC7348] Mahalingam, M., Dutt, D., Duda, K., Agarwal, P., Kreeger,
L., Sridhar, T., Bursell, M., and C. Wright, "Virtual
eXtensible Local Area Network (VXLAN): A Framework for
Overlaying Virtualized Layer 2 Networks over Layer 3
Networks", RFC 7348, DOI 10.17487/RFC7348, August 2014,
<https://www.rfc-editor.org/info/rfc7348>.
[I-D.dukkipati-tcpm-tcp-loss-probe]
Dukkipati, N., Cardwell, N., Cheng, Y., and M. Mathis,
"Tail Loss Probe (TLP): An Algorithm for Fast Recovery of
Tail Losses", draft-dukkipati-tcpm-tcp-loss-probe-01 (work
in progress), February 2013.
[I-D.ietf-nvo3-geneve]
Gross, J., Ganga, I., and T. Sridhar, "Geneve: Generic
Network Virtualization Encapsulation", draft-ietf-
nvo3-geneve-13 (work in progress), March 2019.
Li & Zhou Expires November 25, 2019 [Page 18]
Internet-Draft LOOPS Problem & opportunities May 2019
[I-D.ietf-tcpm-rack]
Cheng, Y., Cardwell, N., Dukkipati, N., and P. Jha, "RACK:
a time-based fast loss detection algorithm for TCP",
draft-ietf-tcpm-rack-05 (work in progress), April 2019.
[]
Filsfils, C., Dukes, D., Previdi, S., Leddy, J.,
Matsushima, S., and d. daniel.voyer@bell.ca, "IPv6 Segment
Routing Header (SRH)", draft-ietf-6man-segment-routing-
header-19 (work in progress), May 2019.
[I-D.cardwell-iccrg-bbr-congestion-control]
Cardwell, N., Cheng, Y., Yeganeh, S., and V. Jacobson,
"BBR Congestion Control", draft-cardwell-iccrg-bbr-
congestion-control-00 (work in progress), July 2017.
[DOI_10.1109_ICDCS.2016.49]
Cai, C., Le, F., Sun, X., Xie, G., Jamjoom, H., and R.
Campbell, "CRONets: Cloud-Routed Overlay Networks", 2016
IEEE 36th International Conference on Distributed
Computing Systems (ICDCS), DOI 10.1109/icdcs.2016.49, June
2016.
[DOI_10.1145_3038912.3052560]
Haq, O., Raja, M., and F. Dogar, "Measuring and Improving
the Reliability of Wide-Area Cloud Paths", Proceedings of
the 26th International Conference on World Wide Web -
WWW '17, DOI 10.1145/3038912.3052560, 2017.
Authors' Addresses
Yizhou Li
Huawei Technologies
101 Software Avenue,
Nanjing 210012
China
Phone: +86-25-56624584
Email: liyizhou@huawei.com
Xingwang Zhou
Huawei Technologies
101 Software Avenue,
Nanjing 210012
China
Email: zhouxingwang@huawei.com
Li & Zhou Expires November 25, 2019 [Page 19]