=Paper=
{{Paper
|id=Vol-3057/paper18.pdf
|storemode=property
|title=Analysis of the Impact of Priority Traffic Control Mechanisms on Network Quality of Service
|pdfUrl=https://ceur-ws.org/Vol-3057/paper18.pdf
|volume=Vol-3057
|authors=Evgenia A. Abramova
}}
==Analysis of the Impact of Priority Traffic Control Mechanisms on Network Quality of Service==
https://ceur-ws.org/Vol-3057/paper18.pdf
Analysis of the Impact of Priority Traffic Control Mechanisms on
Network Quality of Service
Evgenia A. Abramova
The St. Petersburg National Research University of Information Technologies, Mechanics and Optics
(University ITMO), 49 Kronverksky Av, St. Petersburg, 197101, Russia
Abstract
There are many ways of managing traffic in Info Communication networks. The priority ones
are the most relevant for consideration to ensure the proper quality for services that have
requirements for minimum delays or channel bandwidth. However, choosing the most
appropriate way to manage traffic is a task that requires a comprehensive analytical approach.
Within the universal package network, the telephony service is one of many services provided.
In the corporate sector, there has been a noticeable increase in interest in IP telephony services
in recent years, stimulated by some obvious advantages in the form of flexibility and openness
of IP systems, the possibility of using various communication solutions (data, video, voice)
within a single platform, as well as extensive use of wireless technologies. The purpose of the
article is to study the effectiveness of traffic prioritization for Quality of Service (QoS)
management in computer networks. Within the scope of the article, simulation modeling is
carried out in AnyLogic, the behavior of packet delays of different classes is studied when
using various service disciplines. The objects of the study are heterogeneous traffic and its
prioritization disciplines with dynamic priorities, as well as Info - Communication systems and
communication channels. The subject of the study is the disciplines of traffic maintenance with
dynamic priorities within the framework of Info - Communication systems.
Keywords 1
QoS, traffic, FIFO, WFO, low priority, throughput.
1. Introduction
Today, the era of Big Data has come in the field of Info -Communications. The Internet affects
almost all spheres of modern life. Every minute, a huge amount of structured and unstructured data is
transmitted over global networks. There is also a stable annual growth in the number of Internet users.
The main problem of the Big Data era is to optimize resources and increase the efficiency of data
transmission. It should be noted that traffic cannot be considered as a single whole, for various network
services, certain types of network traffic should be selected that meet the efficiency criteria for the
maximum number of simultaneously working users. At the same time, practical limitations in the form
of the width of communication channels and the limited power of network equipment are taken into
account [1].
The work is intended to evaluate the effectiveness of traffic management mechanisms on the global
Internet. This goal is achieved by solving the following tasks:
Determining service quality indicators
Study of traffic varieties and their heterogeneity
Analytical review of service disciplines
Proceedings of VI International Scientific and Practical Conference Distance Learning Technologies (DLT–2021), September 20-22,
2021, Yalta, Crimea
EMAIL: vectra4444@mail.ru
ORCID: 0000- 0002-5637-7427
©️ 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
167
Comparative analysis of existing traffic management mechanisms
In the article, the concept of quality of Service is considered, which implies the solution of two main
tasks:
1. Creating and maintaining the order of receipt of packages
2. Minimizing delays and ensuring positive dynamics of packages transmission
2. Setting the Research Task
The use of data package switching technology can not guarantee high throughput in information
and communication networks. The reason for this is the lack of guarantees for the delivery of the
package. In some applications, the order and delivery intervals do not affect the performance and quality
of user interaction. At the same time, for others, these parameters are fundamental. High traffic service
requirements are not met by the TCP and UDP transport layer protocols because TCP allows some
possible delays in delivery, although it guarantees the correct delivery of packets, and UDP can reduce
delays, but does not provide high-quality traffic service and mechanisms for implementing such. At the
same time, it is necessary to guarantee the delivery of such information like audio, video, and
multimedia in real-time with the minimum possible delay [2]. The primary tasks of the Quality of
Service (QoS) mechanism are:
Creating and maintaining the order of receipt of packages
Minimizing delays and ensuring positive dynamics of packet transmission
Maintaining a high quality of service in the field of IP telephony
The quality of service is defined as "the total effect of the operating parameters of the service, which
determines the degree of user satisfaction with this service". (ITU-T Recommendation E. 800) To be
able to quantify the quality of service in the network, it is necessary to introduce some numerical
parameters. The following parameters are used to evaluate QoS:
Average packet Delivery Delay (IPPacket Transfer Delay). IPTD is defined as the sum of the
delivery times of all packets between the source and the recipient, divided by the number of packets
[3].
Delay variation, jitter (IP Packet Delay Variation). IPDV determines the variability of the delay
in the delivery of consecutive packets.
Packet Loss Ratio (IP PacketLoss Ratio). The IPLR parameter determines the percentage of
packets lost during transmission out of the total of all sent packets.
Packet Error Rate (IP PacketError Ratio). IPER determines the percentage of received packets
that have changed during transmission [4].
3. Maintenance Disciplines
FirstInFirstOut is one of the simplest maintenance disciplines, the essence of which is to process
packets in the same order in which they are initially queued. The use of this discipline in the case of
processing large traffic flows simultaneously leads to the dominance of several of them, which
negatively affects the efficiency of using network resources.
Priority Queueing is a maintenance discipline that involves using a combination of several queues
that are processed using the Taildrop or RandomEarlyDetection buffering disciplines and using the
FirstInFirstOut service discipline within themselves [5]. The distribution of packets to these queues
occurs according to the class of these packets. Then the packets are selected sequentially from these
queues, starting with the queue corresponding to the highest priority. The disadvantage of this
discipline is the fact that packets with low priorities can be processed with significant delays, in some
cases, there may be losses of communication sessions organized using packets with low priorities.
WeightedFairQueueing is a maintenance discipline in which a separate queue using the
FirstInFirstOut discipline is allocated for each traffic class, as well as the allocation of a certain share
of the channel bandwidth for each of these queues [6,7]. The order of servicing these queues, as well as
the capacity of the allocated share of the channel bandwidth, are determined by the packet priorities.
168
4. The Setting of the Experiment
Let us set the following parameters for the study:
Throughput capacity: N = 6 Mbps.
Buffer size: S = 10 Kb.
Priorities WFQ: W1 : W2 =7:1, W1 = 0.875 W2 = 0.125
Table 1
Parameters for Skype messenger and streaming service Twitch
Parameters Skype Twitch
Delay, ms 100 1000
Jitter, ms 50 -
Loss of packets, part 0.001 0.001
The capture of VoIP traffic is carried out as follows: the detection of opened Skype program ports,
detection of the port and address that has the greatest activity during the call, the capture of the traffic
with the filter (Fig. 1).
Figure 1: Screenshot of Skype traffic
VOD traffic is captured as follows: determining the address through which video traffic goes using
the browser's debugging console; capturing traffic with a filter (Fig. 2).
Figure 2: Screenshot of Twitch traffic
If we consider the packet length distribution functions for UDP and TCP traffic, we can see that
UDP packets (Fig. 3) have a fixed maximum size, unlike TCP packets (Fig. 4). In case of loss of a TCP
packet, it will be requested again, and the information will not be lost, unlike UDP - in case of loss of
a packet, it is lost forever[8]. Therefore, it is necessary to transmit smaller UDP packets in order not to
lose large parts of data. As for the distribution function of inter-packet intervals, they are similar in TCP
and UDP and are close to the exponential distribution function.
Figure 3: Packet length distribution function for UDP traffic
169
Figure 4: Packet length distribution function for TCP traffic
4.1. Study of FIFO Discipline
FIFO - an elementary queue without prioritization: each traffic class receives the same amount of
service, taking into account the delay when issuing in the communication channel. Experiments with
throughput values did not lead to obtaining characteristics that meet the specified requirements （Fig.
5, Table 2). This result is associated with a large packet size and a small-time interval between packets:
the specified buffer size and bandwidth are not enough for the selected type of traffic. When trying to
increase the initial parameters, the following results were obtained (Fig. 6, Table 3)
Figure 5: The average waiting time in the model with a capacity of 10 Kbytes and a bandwidth of 6
Kbit/s
Table 2
Parameters of the FIFO model with a capacity of 10 Kbytes and a bandwidth of 6 Kbit/s
Parameters Values
Loading, p 1 +- 2.64Е-5
Chance of loss, π 0.647 +-4.434Е-5
Average waiting time, W, ms 12.986 +- 0.001
Average stay time U, ms 13.365 +- 0.002
Current packet queue length 34
Average I packet queue length, 34.346 +-0.025
170
Picture 6 shows that now the average stay time and the probability of losses are within the normal
range. As a result of the iterative increase of throughput, the model met the QoS criteria better and
better: the delay and the probability of loss decreased.
Figure 6: The average waiting time of the model with a capacity of 10000 Kbytes and a bandwidth of
20.5 Mbit/s
Table 3
Parameters of the model with the capacity of 10000 Kbytes and a bandwidth of 20.5 Mbit/s
Parameters Values
Loading, p 1 +- 1.279Е-4
Chance of loss, π 0.179 +-2.314Е-4
Average waiting time, W, ms 38.407 +- 0.028
Average stay time U, ms 38.57 +- 0.028
Current packet queue length 238
Average I packet queue length 235.75 +-0.423
4.2. Study of PQ Discipline
Priority is set for different traffic classes: low-priority class traffic is transmitted only when there are
no high-priority class packets in the queue. Thus, the best quality of service is provided for the high-
priority class, but the low-priority class is blocked during overloads. It can be seen from the schemes
(Fig. 7,8,9) and tables (Table 4,5) that the change in the bandwidth strongly affects low-priority traffic.
As in the previous case, with variations of the bandwidth, it is not possible to achieve the required
qualities.
Figure 7: AnyLogic model scheme with the capacity of 10 Kbytes and a bandwidth of 6 Kbytes/s
171
Table 4
Table of parameters of the model with a capacity of 10 Kbytes and a bandwidth of 6 Kbytes/s
Parameters Values
Loading, p 1 +- 3.767Е-5
Chance of loss, π 0.66 +-9.465Е-5
Average waiting time, W, ms 0.215 +- 0.02
101.541 +- 0.277
Average stay time U, ms 0.592 +- 0.002
103.331 +- 0.588
Current packet queue length 24
Average I packet queue length 24.783 +-0.036
Figure 8: AnyLogic model scheme with the capacity of 10 Kbytes and a bandwidth of 20.5 Mbit/s
Table 5
Table of parameters of the model with a capacity of 10 Kbytes and a bandwidth of 20.5 Mbit/s
Parameters Values
Loading, p 1 +- 3.754Е-4
Chance of loss, π 0.185 +-5.26Е-4
Average waiting time, W, ms 0.005 +- 2.961Е-4
94.088 +- 0.184
Average stay time U, ms 0.137 +- 8.892Е-4
94.803 +- 0.325
Current packet queue length 143
Average I packet queue length 144.801 +-0.564
Figure 9: The scheme of the dependence of the service time on the bandwidth for PQ
172
4.3. Study of the WFQ Discipline
Weight is set for each traffic. For each cycle of the WFQ operation, packets are transmitted from the
queue of one class, the total size of which is equal to the weight of the class. Setting the weight
guarantees that a class with a higher weight will receive a higher quality of service, and that in
conditions of high load, the class will receive a channel in a finite time[9,10].
Figure 10: AnyLogic model scheme with a capacity of 600 Kbytes and a bandwidth of 600 Kbytes/s
When the weights were varied, the results improved slightly (Pic. 10.11). As in the previous case,
with the variation of the bandwidth, it is not possible to achieve the required qualities. The lowest
bandwidth value was achieved at 23000 bps (Table 6,7). When varying the weights at this throughput
and some values below, it was not possible to achieve significantly better results.
Table 6
Table of parameters of the model with a capacity of 600 Kbytes and a bandwidth of 600 Kbytes/s
Parameters Values
Loading, p 1 +- 4.709Е-4
Chance of loss, π 0.499 +- 0.003
0.891 +- 0.004
Average waiting time, W, ms 0 +- 0
177.249 +- 5.031
Average stay time U, ms 0 +- 0
180.58 +- 5.079
Current packet queue length 354
228
Average I packet queue length 343.702 +- 2.415
184.463 +- 1.867
Figure 11: AnyLogic model scheme with a capacity of Kbytes and a bandwidth of Mbit/s
173
Table 7
Table of parameters of the model with a capacity of 600 Kbytes and a bandwidth of 600 Kbytes/s
Parameters Values
Loading, p 1 +- 0.001
Chance of loss, π 0 +- 0
0.451 +- 0.007
Average waiting time, W, ms 0 +- 0
19.785 +- 1.373
Average stay time U, ms 0 +- 0
21.048 +- 1.455
Current packet queue length 1
151
Average I packet queue length 2.194 +- 0.109
136.205 +- 2.574
The WFQ discipline gives zero latency for high-priority packets, unlike PQ. In both cases, changing
the bandwidth greatly affects low-priority traffic. (Fig. 12,13)
Figure 12: Scheme of changes in the probability of bandwidth losses for FIFO, PQ, and WFQ
Figure 13: Scheme of change of service time due to the bandwidth for FIFO and PQ
174
5. Conclusion
As a result of the work, the following results were obtained:
1. When the bandwidth of the channel increases, the characteristics decrease to a certain threshold,
after which the bandwidth increase has no effect
2. The FIFO Service Discipline does not provide traffic management mechanisms.
3. The PQ Service Discipline provides an elementary control mechanism that cannot handle
overload cases.
4. The WFQ Service Discipline provides a traffic management mechanism by assigning weights
to the classes into which traffic is divided; it is more flexible than the previous two, and it copes
with congestion better than PQ.
5. The initial configuration (6 Mbit/s bandwidth and 600 Kbyte/s buffer) is unacceptable when
applying any Service Discipline in the case of simultaneous use of Skype and Twitch.
6. If it is necessary to separate the quality of service and the minimum bandwidth, WFQ will be a
suitable option, otherwise, FIFO will do.
6. References
[1] V.A. Bogatyrev, A.V. Bogatyrev, S.V. Bogatyrev, Redundant Servicing of a Flow of
Heterogeneous Requests Critical to the Total Waiting Time During the Multi-path Passage of a
Sequence of Info-Communication Nodes. Lecture Notes in Computer Science (including subseries
Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). 2020. Vol. 12563.
pp. 100-112. DOI 10.1007/978-3-030-66471-89.
[2] V.A. Bogatyrev, S.V. Bogatyrev, A.N. Derkach, Timeliness of the Reserved Maintenance by
Duplicated Computers of Heterogeneous Delay-Critical Stream. CEUR Workshop Proceedings.
2019. Vol. 2522. pp. 26-36.
[3] V.A. Bogatyrev, S.V. Bogatyrev, A.V. Bogatyrev, Redundant multi-path service of a flow
heterogeneous in delay criticality with defined node passage paths. Journal of Physics: Conference
Series, Volume 1864, 13th Multiconference on Control Problems (MCCP 2020) 6-8 October 2020,
Saint Petersburg, Russia 2021 J. Phys.: Conf. Ser. 1864 012094 - 2021, Vol. 1864, 012094, No. 1,
pp. 012094. DOI 10.1088/1742-6596/1864/1/012094.
[4] M. Rouached, W. Fdhila, and C. Godart, Web services compositions modeling and choreographies
analysis, International Journal of Web Services Research, vol. 7, no. 2, pp. 87–110,
2020.DOI: 10.4018/jwsr.2010040105
[5] L. Li, S. Li, and S. Zhao, QoS-Aware scheduling of services-oriented internet of things, IEEE
Transactions on Industrial Informatics, vol. 10, no. 2, pp. 1497–1507, 2014.DOI:
10.1109/TII.2014.2306782
[6] W. Sugeng, J.E. Istiyanto, K. Mustofa, and A. Ashari, “The Impact of QoS Changes
towards Network Performance”, International Journal Computer Networks and
Communications Security, vol. 3, no. 3, pp. 48-53, February 2015. DOI: 10.34128/jsi.v5i2.191
[7] D. Mistry, P. Modi, K. Deokule, A. Patel, H. Patki, and O. Abuzaghleh, “Network
traffic measurement and analysis”, 2016 IEEE Long Island Systems, Applications and
Technology Conference (LISAT), Farmingdale, NY, pp. 1-7, 2016.
DOI: 10.1109/LISAT.2016.7494141
[8] R. S. Matos, P. R. M. Maciel, and R. M. A. Silva, QoS-driven optimization of composite web
services: An approach based on GRASP and analytical models, International Journal of Web and
Grid Services, vol. 9, no. 3, pp. 304–321, 2013.
[9] Ga-Won Lee, Sung-Young Lee, and Eui-Nam Huh “Congestion Prediction
Modeling for Quality of Service Improvement in Wireless Sensor Networks”,
Sensors, 14, 7857-7880, 2014 DOI: 10.3390/s140507857
[10] E.A Abramova, I.V. Kalinin, D.A. Mezentsev, L.A. Muravyeva-Vitkovskaya, Models and research
methods of heterogeneous traffic control processes in information and communication systems //
20th International Multidisciplinary Scientific GeoConference Surveying Geology and Mining
Ecology Management, SGEM-2020 - 2020, No. 2.1, pp. 223-230.
175