CC BY
25
COMPARISON OF ALGORITHMS FOR CONSTRUCTING QUEUES WHILE ENSURING QUALITY OF SERVICE
Sergey V. Ryzhikh,
PSUTI, Samara, Russia, rrserg@mail.ru
Lyubov A. Marykova,
PSUTI, Samara, Russia, l.marykova@mail.ru
Mikhail V. Marykov,
"Netcracker Technology", Samara, Russia,
mikhail.marykov@gmail.com
Boris Y. Lihtsinder,
PSUTI, Samara, Russia, lixt@samtel.ru
Clitheroe Sean,
"SilkSmith", United Kingdom, blacker.mini@gmail.com
Keywords: queuing mechanisms, PQ, SCFQ, sfq, WF2Q, WFQ, LLQ, loss, unilateral latency.
In existing data communication networks an overload of channels may happen which leads to loss of traffic. To smooth short-term overloads, the internal buffer memory of the switch is used. It is necessary for temporary storage of data frames in cases where they can not be immediately transmitted to the output port. The buffer is used for smoothing short-term traffic pulsations. Even if the traffic is well balanced and the performance of the port processors as well as other processing elements of the switch is sufficient to transmit average traffic values, this does not guarantee that their performance will be sufficient for peak loads. For example, traffic can flow for several tens of milliseconds simultaneously to all inputs of the switch, preventing it from transmitting received frames to the output ports. If the average value of traffic intensity is repeatedly exceeded by a multiple (and for local networks the values of the traffic pulsation coefficient in the range of 50-100 are often encountered), there may be frame losses. One of the methods to prevent this to queue them in the memory buffer. The frame, depending on its priority, falls into a certain queue and after that a specially selected mechanism determines in which order frames from different queues will be sent to the output port. For this, queue serving mechanisms such as FIFO, LLQ, PQ, and SCFQ are used.
This article includes an overview of these mechanisms and compares the various types of traffic such as voice traffic with DSCP 46 quality of service, PHP EF, Premium class, Video traffic with DSCP 10 quality of service, PHP AF10 class Gold, critical data with quality service DSCP 12, AF12 class Gold, and non-critical data with the quality of service DSCP 0, class Best-Efforts. The efficiency of these mechanisms is estimated by the following network characteristics: losses, the difference in delays is IPDV (IP Packet Delay Variation), one-way delay. Also this article gives recommendations for use of mechanisms in existing or projected communication networks.
Information about authors:
Sergey V. Ryzhikh, Graduate Student of the Volga State University of Telecommunications and Informatics (PSUTI), Samara, Russia
Lyubov A. Marykova, Associate Professor at the Department of Communication Systems, PSUTI, Samara, Russia
Mikhail V. Marykov, Business Analyst, "Netcracker Technology", Samara, Russia,
Boris Ya. Lihtsinder, Professor at the Department of MSIB, PSUTI, Samara, Russia
Sean Clitheroe, Engineer of Telecommunication, "SilkSmith" company, United Kingdom
Для цитирования:
Рыжих С.В., Марыкова Л.А., Марыков М.В., Лихтциндер Б.Я., Клитероу Ш. Cравнение алгоритмов построения очередей при обеспечении качества обслуживания // T-Comm: Телекоммуникации и транспорт. 2017. Том 11. №11. С. 74-79.
For citation:
Ryzhikh S.V., Marykova L.A., Marykov M.V., Lihtsinder B.Y., Clitheroe S. (2017). Comparison of algorithms for constructing queues while ensuring quality of service. T-Comm, vol. 11, no.11, pр. 74-79.
T-Comm ^м 11. #11-2017
Introduction
Sometimes in an existing data network, a channel is overloaded and causes loss of traffic. To smooth short-term overloads, systems use an internal buffer memory switch. It is necessary to temporarily store the data frames in those eases where they ean not be immediately transferred to the output port.
The buffer is used for smoothing short-term fluctuations of traffic. Alter all, even if the traffie is well balanced and the performance of processors ports, as well as other processing elements Of the switch is sufficient to transmit the average values of traffic, there is no guarantee that their performance will be enough at peak loads. For example, the traffic can be within a lew tens of milliseconds acting simultaneously on ail the inputs ports of the switch, not giving it the opportunity to transmit a frame as it is received at the output port. In the short-term, repeatedly exceeding the mean value of traffic intensity (for local area netw orks this commonly occurs in values of ripple traffic in the range 50-100) possible causing frame loss. One way to cope with this is to form a queue in the buffer memory. A frame, depending on its priority, is placed in a specific queue and then a specially selected mechanism determines the sequence in which frames of different queues are forwarded to the output port. For this purpose queuing mechanisms, such as f'Q. SCFQ, SFQ. IVF2Q. WFQ, LLQ can be used.
This article provides an overview of these mechanisms and a comparison with the maintenance of different types of traffic, such as prioritized traffic with QoS DSCP 46, PHP EF, class Premium, and non-prioritized traffic with the service class Best-Efforts. The effectiveness of these mechanisms is evaluated according to the following network characteristics: loss; difference in delays — 1PDV (IP Packet Delay Variation): oneway delay. Recommendations are then made on the use of mechanisms in existing or planned networks.
Currently, data networks appear to be doubling in traffic every half year. This is caused by an increase in the services offered to users. Such rapid growth commonly leads to overloads in the existing network. Overloads happens on the output port of the switch.
The peculiarity of the Ethernet switches is that lor temporary storage of frames, each switch port has an internal buffer memory. If the communication channel is free, then the frame is transmitted immediately to the destination. If this is not possible, then the switch puts frame in temporary storage within the buffer memory, forming a queue. The procedure for transfer via the output port of queued packets is based on their priorities and is determined by a queuing mechanism, which allows management of the network bandwidth in the case of overloads.
This article contains a study of how to apply queuing mechanisms to provide quality of service. The main characteristics of the network thus will be: available bandwidth, one-way delay, the difference in the delay and the percentage of lost packets.
Let us consider a queue like an ordinary traffic stream consisting of a finite number of packets arriving for processing in the switch. Assume that each packet is processed in a constant lime t . We divide the entire reporting interval N into equal intervals, with the duration r . The order number of the intervals
is denoted by /. Due to the ordinary flow, during each / — the time interval an integer number of packets is received mi (r)
(/=0,1,2...). Consequently, the number of packets received during time r, is a discrete random variable with the mathematical expectation m (¡-). During each t- time interval in
the router there are i/,(r) packets waiting in the queue. Queue
length is also a discrete random variable with the mathematical expectation q
We denote by q. (r) random-number sequence shifted with
respect to qt{j) at j intervals r.
For any system with a single queue it is fair to say that a balance occurs between the incoming and processed packets.
qi{T) = q,_x{T) + ml(T)-$l{T). (1)
c>- = | 0 & 1 == 0 (2) I otherwise
Pay attention to some obvious features S,(t)'
= ^(^(r) = »;,(!-); (3)
Consider the formula (1) applied to the network device with a finite length of the queue . Based on this we obtain
equilibrium:
fO if <7M + ill > d,
Also we can calculate the length of the queue taking into account dropped packets caused by the queue overflow rt
r i 9; if <!i < <7mas U™ 'f <7, £
And, consequently, the number of dropped packets caused by the queue overflow I.:
/ JO if
</ H, ^ <7™,
The network device can allocate resources sufficient to provide a certain intensity of outgoing traffic through the use of different functional implementations of the policy. When it comes to pass all traffic through an overloaded network segment (which implies the presence of large accumulated queues), this feature ean be implemented using various queuing mechanisms, such as PQ, SCFQ, SFQ, WF2Q, WFQ, LLQ.
The practical part.
Considering implementation and queue management in practice. Figure I illustrates the network diagram for the investigation
■Èfl
Visual analysis of the graphs of the studied algorithms allows us to draw some conclusions:
- The smallest delay occurs when using the algorithm LLQ and the maximum with PQ;
- The difference in latency for the LLQ algorithm is characterized by a more uniform value at length packet 151S.
Evaluation of unilateral latency
The unilateral latency parameter is del med as time {/, -tt) between the two events - input a package into the entry point of
a network at the moment f, and output the packet from the output point of the network at the moment t2, where
(?2-*ä and & ~tt)itmK
In general, the unilateral latency parameter is determined as the deliver time for the packet between the source and destination for all packages — including successfully transmitted as well as packets affected by errors.
HI' >1
0n»w»f Gtftï to tf MuHf
Ï. I"
i
Fig. 7, Unilateral latency parameter for packet ¡ength 1518
Fig. 9. Unilateral latency parameter for packet length 64
flvï IkK-iUH Dvl*u for CI traftir 'ri r"'l - idALt
K't Ow-uvj Detail far ET trWÏ ü 'II Pn«4) - iclwil
Fig, 8. Unilateral iatcncy parameter for algorithm LLQ for packet length 1518
:•»: 1H 1«
Fig. 10. Unilateral latency parameter for algorithm LLQ for packet length 64
Visual analysis of the graphs allows us to draw the following conclusions:
- Unilateral latency is lowest when using the LLQ algorithm and highest for WF2Q;
- Unilateral latency values depend on the length of the packet;
Conclusions
Providing the required quality of service in networks can certainly be reached by the direct method, through the provision of guaranteed bandwidth or improving the performance of network deviccs. However, the most appropriate way is the use of methods discussed in this article, which provide the required service quality while making efficient use of the network resources for a wide range of different applications, including the most critical audio and video applications. Thus, based on the data, all the mechanisms showed acceptable results. The choice of algorithm depends on the specifics
of the traffic in the network. In this article, the LLQ algorithm has showTt the best results.
In our next research we will compare performance of these algorithms in the construction of a larger number of queues and the use of different classes of service.
References
1. The «.*, Kevin Fall, 2008 «The V1NT/CONSER Project» Xerox PARC , 2008. p. 288 .
2. Knel QoS Course Administration Guide , M&T Books , 2005 C, pp. 155-157,
3. CCiE - Cisco Certified internetwork Expert. John Schwartz, Todd Lemmle, 2008. p. 510
4. RFC 1122, RFC 2475 , RFC 4594 , RFC 3247, RFC 2474, RFC 2475 . RFC 1633, RFC 5290, RFC 3246.
5. Standard of protocol ¡EEE Sid 802. Iaq™-2012,> The Institute of Electrical and Electronics Engineers, Inc. Chapter 4, p. 145.
CРАВНЕНИЕ АЛГОРИТМОВ ПОСТРОЕНИЯ ОЧЕРЕДЕЙ ПРИ ОБЕСПЕЧЕНИИ КАЧЕСТВА ОБСЛУЖИВАНИЯ
Рыжих Сергей Вячеславович, ПГУТИ, Самара, Россия, rrserg@mail.ru Марыкова Любовь Анатольевна, ПГУТИ, Самара, Россия, l.marykova@mail.ru Марыков Михаил Валерьевич, ООО "НетКрэкер", Самара, Россия, mikhail.marykov@gmail.com Лихтциндер Борис Яковлевич, ПГУТИ, Самара, Россия, lixt@samtel.ru Клитероу Шон, компания "SilkSmith", Кембридж, Великобритания, blacker.mini@gmail.com
Aннотация
В существующих сетях передачи данных порой происходит перегрузка каналов, что влечет потерю трафика. Для сглаживания кратковременных перегрузок используется внутренняя буферная память коммутатора. Она предназначена для временного хранения кадров данных в тех случаях, когда их невозможно немедленно передать на выходной порт.
Буфер предназначен для сглаживания кратковременных пульсаций трафика. Если трафик хорошо сбалансирован и производительность процессоров портов, а также других обрабатывающих элементов коммутатора достаточна для передачи средних значений графика, это не гарантирует, что их производительности хватит при пиковых значениях нагрузок. При кратковременном многократном превышении среднего значения интенсивности трафика (а для локальных сетей часто встречаются значения коэффициента пульсации трафика в диапазоне 50-100) возможны потери кадров. Одним из методов борьбы с этим служат очереди в буфере памяти.
Кадр, в зависимости от его приоритета, попадает в определенную очередь, и после этого, специально выбранный механизм определяет, в какой последовательности кадры из разных очередей будут пересылаться в выходной порт. Для этого используются механизмы обслуживания очередей, такие как FIFO, LLQ, PQ и SCFQ.
В этой статье содержится обзор данных механизмов и проводится сравнение при обслуживании разных типов трафика, таких как голосовой трафик, трафик с качеством обслуживания DSCP 46, PHP EF, видео трафик с качеством обслуживания DSCP 10, PHP AFI0, критичные данные с качеством обслуживания DSCP 12, AFI2, некритичные данные с качеством обслуживания DSCP 0, класс Best-Efforts. Эффективность этих механизмов оценивается по следующим сетевыми характеристиками: потери; разница в задержках - IPDV (IP Packet Delay Variation); односторонняя задержка. Даются рекомендации по использованию механизмов в существующих или проектируемых сетях связи.
Ключевые слова: механизмы обслуживания очередей, PQ, SCFQ, SFQ, WF2Q, WFQ , LLQ, потери, односторонняя задержка.
Литература
1. The ns, Kevin Fall, 2008 "The VINT/CONSER Project" Xerox PARC, 2008, 288 c.
2. Knet QoS Course Administration Guide, M&T Books, 2005. С. 155-157.
3. CCIE - Cisco Certified Internetwork Expert. John Schwartz, Todd Lemmle, 2008. С. 510.
4. RFC 1122, RFC 2475, RFC 4594, RFC 3247, RFC 2474, RFC 2475, RFC 1633, RFC 5290, RFC 3246.
5. Standard of protocol IEEE Std 802.laq™-20l2.' The Institute of Electrical and Electronics Engineers, Inc. Chapter 4, p. 145.
Информация об авторах:
Рыжих Сергей Вячеславович, ФГБОУ ВО ПГУТИ, соискатель, Самара, Ро^ия Марыкова Любовь Анатольевна, ФГБОУ ВО ПГУТИ, доцент, Самара, Россия Марыков Михаил Валерьевич, ООО "НетКрэкер", бизнес-аналитик, Самара, Россия Лихтциндер Борис Яковлевич, ФГБОУ ВО ПГУТИ, д.т.н., профессор, Самара, Россия Клитероу Шон, "SilkSmith", инженер, Великобритания
