A Method for Exploring Latency Correlation in Multipath Networks
draft-zhou-pce-latency-00
| Document | Type | Active Internet-Draft (individual) | |
|---|---|---|---|
| Authors | Ying Zhou , Mingzhen Wu | ||
| Last updated | 2024-04-17 | ||
| RFC stream | (None) | ||
| Intended RFC status | (None) | ||
| Formats | |||
| Stream | Stream state | (No stream defined) | |
| Consensus boilerplate | Unknown | ||
| RFC Editor Note | (None) | ||
| IESG | IESG state | I-D Exists | |
| Telechat date | (None) | ||
| Responsible AD | (None) | ||
| Send notices to | (None) |
draft-zhou-pce-latency-00
Independent Submission Zhou Ying
Internet Draft Wu Mingzhen
Intended status: Informational
Expires: September 10, 2024
Beijing Jiaotong University
April 18, 2024
A Method for Exploring Latency Correlation in Multipath Networks
draft-zhou-pce-latency-00
Abstract
The exploration of latency correlation patterns is of great
significance for characterizing network states. However, existing
measurement approaches have to confront multiple challenges in
detecting latency correlation factors, such as probing speed,
routing hops and geographical locations. In this paper, we conduct
three tasks to handle these issues. The first is to construct
relative latency measurement strategies for individual machines
without any time synchronization and control plane requirements. The
probing speed has been increased by 14.3% in the same hardware
conditions. In 4G/5G heterogeneous edges enabled by different mobile
operators, worldwide 5003 target servers are selected to acquire
more than 1TB network datasets. The second is to reveal the
potential modes between latency and routing hops. Surprisingly,
91.2% available targets present non-positive characteristics in
extreme cases. The third is to analyze the fine-grained relationship
between latency and geographical locations. We found that the
significance of mean backward receiving delay is higher than other
parameters, with a maximum of 33%. Finally, we also made
optimizations regarding latency compression and time accuracy in
multipath networks. The experimental data have been released to an
open-source community for further investigations.
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), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other documents
Zhou, et al. Expires September 10, 2024 [Page 1]
Internet-Draft Latency Correlation April 2024
at any time. It is inappropriate to use Internet-Drafts as
reference material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
This Internet-Draft will expire on September 10, 2024.
Zhou, et al. Expires September 10, 2024 [Page 2]
Internet-Draft Latency Correlation April 2024
Copyright Notice
Copyright (c) 2024 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 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....................................................... 4
2. Scheme Design...................................................... 4
3. Transmission Optimization.......................................... 7
3.1. Path Selection ............................................... 7
3.2. Speed Improvement ............................................ 7
3.3. Accuracy Improvement ......................................... 8
4. Conclusion......................................................... 8
5. Unresolved Issues.................................................. 8
6. Security Considerations............................................ 8
7. References......................................................... 9
8. Acknowledgments.................................................... 9
Zhou, et al. Expires September 10, 2024 [Page 3]
Internet-Draft Latency Correlation April 2024
1. Introduction
With the increasing adoption of new network devices and
communication technologies, understanding the changes in network
Quality of Service (QoS) is crucial. Among various QoS metrics, one-
way latency has gained significant attention for ensuring QoS
provisioning, verifying Service Level Agreements (SLAs), and
detecting network anomalies. However, exploring the variation
patterns of one-way delay (OWD) faces several challenges, including
unsynchronized device clocks, difficulty in controlling server
targets, and security concerns. To address these challenges, we
propose an end-to-end relative latency measurement scheme capable of
probing 5003 global servers and gathering 1TB datasets in
heterogeneous network environments.
The relationship between relative one-way latency, routing hops and
geographical location has been established. The contributions of
this paper are as follows:
• A new network measurement strategy has been constructed: A
relative OWD measurement is designed to avoid the requirements of
time synchronisation, software-defined networks (SDN), operating
licences or other limitations for latency parameter measurement.
• Correlation analysis of latency and routing hop count: Through the
measured data, it is revealed that there is a non-positive
correlation between relative OWD and routing hop count in 91.2% of
measurement points in extreme cases.
• Correlation analysis of latency and geographical location: The
experiment displays a latency-geographic distance relationship
indicator. The correlation between average backward receiving delay
and distance is significantly higher than other average latency
parameters, with a maximum increase in significance of 33%.
2. Scheme Design
In this section, we explain the mathematical model of measurement
strategies, which includes measuring the structure and sending
method of the message, obtaining the target, data processing, and
keyframe algorithms. Additionally, a fourth timestamp, TW ireshark,
is acquired by utilising network packet capture software.
⚫ Mathematical Model
This research utilise NTP as the primary measurement protocol.
The data packet comprises three essential timestamps, namely
TOriginate, TReceive and TT ransmit.
Zhou, et al. Expires September 10, 2024 [Page 4]
Internet-Draft Latency Correlation April 2024
⚫ Message Structure
Measurement message structure design: In order to gain
comprehensive network measurement performance, more targeted
modifications must be made to network packets.
An innovative network measurement message format was developed
based on NTP [20].
+-------------------------------------------------------------+
| 4-bit | 4-bit | 8-bit Type | 16-bit Total Length |
| version | IHL | Of Service | |
+-------------------------------------------------------------+
| 16-bit Identification |3-bit | 16-bit Total Length |
| |Flags | |
+-------------------------------------------------------------+
| 8-bit Time To | 8-bit | 16-bit Header Checksum |
| Live TTL=1…30 | Protocol | |
+-------------------------------------------------------------+
|32-bit Source Address Source Address={5G IP address(Telecom);|
| 4G IP address(Mobile) ; 4G IP address(Telecom)} |
+-------------------------------------------------------------+
| 32-bit Destination Address |
| Destination Address={NTP server IP addresses worldwide} |
+-------------------------------------------------------------+
| 16-bit Source Port | 16-bit Destination Port |
| S Port=49152…65535 | |
+-------------------------------------------------------------+
| 16-bit Length | 16-bit Checksum |
| | |
+-------------------------------------------------------------+
| NTP message |
+-------------------------------------------------------------+
Fig. 1. Network packet structure for measurement, modified port
number, source address, destination address, etc
As shown in Fig. 1, the image represents the ip and user datagram
protocol (UDP) header of the measurement message. The red field
has been modified, and we have changed the Time To Live (TTL) to
1-30 to detect forward routing information. In addition, based on
heterogeneous network environments, inorder to bind network
measurement messages to different networks and send multiple
measurement points, we modified the source and destination
addresses of the message. We also made modifications to the
source port number, gradually increasing it from 49152 to 65535.
Finally, we mainly extracted three timestamps from the NTP
message, namely the original timestamp, the receive timestamp,
and the transmit timestamp.
Zhou, et al. Expires September 10, 2024 [Page 5]
Internet-Draft Latency Correlation April 2024
⚫ Obtain Measurement Targets
We use the multi-location DNS resolution method. Alibaba and
Tencent Cloud servers in Beijing, Shanghai and Shenzhen are used
for DNS resolution of measurement targets. In this process, the
server queries the local name server, root name server,top-level
name resolution server and the right name server.Through this
method, the data obtained from different locations are not the
same. After deduplication and cleaning of the data, this study
can collect as many NTP servers as possible to ensure the
reliability and effectiveness of the network measurement dataset.
⚫ Relative OWD Measurement Method
The construction timestamp interval is too large: In order to
obtain meaningful relative latency parameters, it is required
that three packets transmitted through different networks be sent
simultaneously, and the interval between the construction times
of the three packets should be as small as possible. Since the
message transmission method based on the ntplib library is
difficult to meet the time interval requirements of the
experiment,1‘ it is urgent to improve and optimize the network
measurement method.
Transmitting and receiving rounds: To overcome this difficulty,
the present work employs two distinct approaches for message
transmission, one of which is the transmitting and receiving
rounds method. After the data message is transmitted, the current
contracting process won’t finish until the reply data packet is
received or the predefined waiting time limit is surpassed, at
which point the next contracting process will begin.
Parallel receiving and sending: We use a parallel receiving and
sending method. This method can quickly gather several actual and
available data packets. However, since the disordered method of
sending and receiving may cause data packet dislocation, it is
necessary to require secondary data processing and cleaning
before it can be used for data analysis, which is not ideal for
online and real-time data processing.
Message sending method that meets time accuracy: This study has
developed a raw socket sending method for parallel sending and
receiving packets. Experimental comparisons show that the socket
sending method has a smaller packet time interval and better
timestamp accuracy. Through this method, messages from three
different networks can be created and sent in 0.001 ms,
satisfying the experiment’s time period.
⚫ Keyframe Algorithms
Zhou, et al. Expires September 10, 2024 [Page 6]
Internet-Draft Latency Correlation April 2024
In a small round of packet sending, a total of 30 network
measurement messages were sent. The corresponding response
message only needs to extract one packet with the smallest TTL
value to obtain all network performance measurement information
in this round of message sending, which is called a ”keyframe”.
The program blocks the dataset in order of network type, original
address, and destination address, so that we can conveniently
obtain the data of each individual network for each measurement
target. Then, by sorting the time and port number information,
the keyframe data for each measurement round can be extracted. It
contains all the information required for network measurement in
this experiment, including the relative total delay, OWD, one-way
routing hops, etc.
3. Transmission Optimization
This section introduces network optimization based on measurement
data, including reducing total latency, improving message sending
rate, and improving time accuracy.
3.1. Path Selection
Through the network measurement method proposed in this draft, the
one-way delay performance of multiple networks for a single device
can be obtained, and the delay performance of different networks can
be compared for different communication targets. Some networks have
good one-way latency performance, which provides optimization space
for path selection. By simultaneously using multiple networks to
connect to the destination device and filtering out networks with
better forward latency and networks with better backward latency,
packets can be sent and received through these two networks. This
allows for the simultaneous use of the network with the shortest
forward latency and the network with the shortest backward latency,
thereby compressing the overall round-trip latency.
3.2. Speed Improvement
By using the network measurement message format mentioned in Fig. 1,
the separation of message sending and receiving can be achieved,
thereby accelerating the collection speed of measurement data. In
the normal one-time message sending process, both the sender and
receiver must establish sockets to listen for message replies.
However, if the port number is used as an identifier to identify the
order of message sending, there is no need to listen for the message
process.
Zhou, et al. Expires September 10, 2024 [Page 7]
Internet-Draft Latency Correlation April 2024
3.3. Accuracy Improvement
The network measurement method using the NTP protocol has a benefit,
as it allows for device time synchronization during network
measurements, which greatly improves the time accuracy of wide area
network devices. At the same time, there is also room for
optimization in multiple network measurements, such as using the
least squares method to optimize time synchronization differences.
4. Conclusion
We aim to explore the changes in large-scale unidirectional latency
and propose a network measurement strategy that does not require SDN
architecture, time synchronization, and extra equipment. Through
analysis of measurement data, our experimental results has verified
that latency at some targeted sites and route hops meets the
negative correlation. It is speculated that the real-time change of
the routing table makes the packets forward from the path with a
lower latency.
Although route hops is large, the one-way latency is low.
In addition, the correlation analysis between geographical location
and latency shows that compared to RTT, backward receiving delay is
more significant in the calculation, fitting, and prediction tasks
of the correlation due to its lower volatility. Among them, due to
the unknowability of the lowest latency, in scenarios with high
real-time requirements, the average backward receiving delay is the
optimal choice.
In this measurement process, we completed the optimal network
selection under asymmetric paths, reduced the total delay and
fluctuation of the time synchronisation difference.
5. Unresolved Issues
The data can be used for simulation experiments to establish traffic
delay models for different routing jump numbers in different
geographical locations.
The measurement method can also be used to optimize multiple time-
related network service functions.
6. Security Considerations
This document does not contain any security considerations.
Zhou, et al. Expires September 10, 2024 [Page 8]
Internet-Draft Latency Correlation April 2024
7. References
PCEP Extensions for Signaling Multipath Information [draft-ietf-pce-
multipath-11].
PCEP extensions for p2mp sr policy [draft-ietf-pce-sr-p2mp-policy-
05]
Inter Stateful Path Computation Element (PCE) Communication
Procedures. [draft-ietf-pce-state-sync-07]
8. Acknowledgments
Many thanks to all who discussed this with us in DINRG in 2023 and
2024.
This document was prepared using 2-Word-v2.0.template.dot.
Authors’ Addresses
Ying Zhou
Beijing Jiaotong University (BJTU)
Beijing, 100044, P.R.China
Email: 22110019@bjtu.edu.cn
Mingzhen Wu
Beijing Jiaotong University (BJTU)
Beijing, 100044, P.R.China
Email: 23125063@bjtu.edu.cn
Zhou, et al. Expires September 10, 2024 [Page 9]