The Effects of Different Congestion Management Algorithms over Voip Performance

This paper presents one of the features of DS (Differentiated Services) architecture, namely the queuing or congestion management. Packets can be placed into separate buffer queues, on the basis of the DS value. Several forwarding policies can be used to favor high priority packets in different ways. The major reason for queuing is that the router must hold the packet in its memory while the outgoing interface is busy with sending another packet. The main goal is to compare the performance of the following queuing mechanisms using a laboratory environment: FIFO (First-In First-Out), CQ (Custom Queuing), PQ (Priority Queuing), WFQ (Weighted Fair Queuing), CBWFQ (Class Based Weighted Fair Queuing) and LLQ (Low Latency Queuing). The research is empirical and qualitative, the results are useful both in infocommunication and in education. Keywords—CBWFQ; congestion; CQ; FIFO; LLQ; Pagent; PQ; queuing; WFQ


INTRODUCTION
At the beginning computer networks were designed mainly for data transfer such as FTP and email, where delay was considered to be unimportant.In most cases the delivery service was effective, and the TCP protocol dealt with data losses.As the multimedia applications became popular (voice transfer, video conferences), separate telephone and video communication networks were set up (see Fig. 1).Nowadays, office and company networks are transformed into one converged network (see Fig. 2), in which the same network infrastructure is used to ensure all the requested services [1].
Although converged networks have many advantages, there are some disadvantages too, namely the competition for network resources (buffers of routers), which leads to congestion.Delay in delivering the packets, jitter, loss of packets are consequences of congestion.Different applications show different sensitivity to these issues.For example, FTP is not impacted by delay and jitter, whereas the multimedia applications (e.g.interactive voice) are very sensitive to them and the loss of packets too [2].Quality of Service (QoS) was introduced to handle this problem, and it is able to provide different priority to different applications, users, or data flows, or to guarantee a certain level of performance to a data flow [3], [4].

Fig. 2. A converged infocommunication network
In accordance with the QoS requirements and recommendations for the interactive voice traffic packet loss should be no more than 1%, one-way latency should be not exceed 150ms and the average one-way jitter should be targeted at less than 30ms [5].
In the IP header there are some fields which can be used to make distinction between the packets of different applications, for example the Type of Service (ToS) field [6].Different technics are used for congestion management (PQ, CQ, WFQ, CBWFQ and LLQ).Congestion avoidance (WRED), traffic shaping and traffic policing are also used by the QoS technology in order to control data traffic [7].This article focuses on the most important component, namely the congestion management.
Our purpose is to analyze and evaluate the efficiency of these congestion management algorithms using a laboratory www.ijacsa.thesai.orgenvironment.This paper examines the following methods: FIFO, PQ, CQ, CBWFQ, WFQ and LLQ.It is important to note that these algorithms have real effect only in the case of congestion.
The network topology for the performance evaluation is identical to the one used in former articles (see e.g.[8]- [10]).This paper continues to study the queuing technologies for congestion management.In [8] and [11] the authors considered three algorithms: FIFO, PQ and WFQ.The conclusion was that WFQ is the most efficient for multimedia applications.In addition to these three new algorithms were investigated: CQ, CBWFQ and LLQ.The main result of this paper is that for multimedia applications (mainly for voice transfer) LLQ is more efficient then WFQ.
The detailed description of the algorithms has been discussed in several papers already (see e.g.[12]- [14]).

II.
THE MEASUREMENT ENVIRONMENT The measurement environment network topology is shown on Fig. 3, which was built at the network laboratory of the Faculty of Informatics, University of Debrecen.The measurement environment consists of three routers (two Cisco 2811 type routers and one Cisco 2801 router) and two switches.The routers are connected with a point-to-point link, having the speed of 128000 cycles per second.The rest of the network devices are connected with 10 Mbps Ethernet links.The part between the two routers is actually a narrow cross-section where congestion can happen.For this reason the congestion management algorithms are activated in this area (see [15]- [16]).
The Cisco IOS 12.4 operating system was running on the R1 and R2 routers, represented on Fig. 3.The TrafGen router was responsible for the functioning of the communication endpoints.This was used to generate the traffic, and the operating system run on TrafGen was c2801-tpgen+ipbasemz.PAGENT.4.3.0 [17], which enabled the traffic generation, attached timestamps to the outgoing packages, and performed the statistical analysis based on the rate of incoming packets.
In order to distinguish between the generated and incoming traffic, two Cisco 2960 switches were used (SW1, SW2).These created two Virtual LANs, namely VLAN 10 and VLAN 20 [18].A serial connection was created between R1 and R2, which simulated a slow WAN.Three types of traffic were generated, similarly to the previous papers: an FTP, Video and VoIP traffics.In the next section the traffic generation code is shown used by TrafGen router.

A. The traffic generation
The following code was used for traffic generation [19]:

B. The implementation of congestion management algorithms
The part between the R1 and R2 routers is actually a narrow cross-section where congestion can happen.For this reason the congestion management algorithms were activated in this area (between the R1' S 0/1/1 and R2' S 0/1/0 interfaces).These codes (for FIFO, PQ, CQ, WFQ, CBWFQ and LLQ) can be found in APPENDIX.

III. MEASUREMENT RESULTS
As in the previous works [9]- [10], the length of the measurement time was 5 minutes in each case.The measurements were recorded in every second.Easy to observe in the traffic generation code, that the generated voice traffic average was 513 packets per second.As in previous articles (see e.g.[8]- [11]) the following areas were examined: packet loss, end-to-end delay and jitter (delay variation).
Concerning the packet loss of voice packets (see Fig. 4) LLQ and PQ algorithms have proven to be most effective, followed by the CBWFQ.It can be observed that while the previous works based on simulation results concluded that WFQ was the best congestion management algorithm, our measurement results showed, that the WFQ performance was behind the performance of LLQ, PQ and CBWFQ.The CQ has the next poor algorithm performance, while the least efficient queuing scheduler was the FIFO.  5 shows the same content as Fig. 4, except that the former does not include the efficiency representation of the two least efficient congestion management algorithm.Thus it is prominently observable the difference of performance of PQ, WFQ, CBWFQ and LLQ in packet loss.Easy to see, that in the case of LLQ and PQ there was no packet loss due to the absolute priority queue, in which the real-time voice was classified.Subsequently, CBWFQ performance was the most effective, and finally the WFQ's.With respect of voice packet delay (see Fig. 6 and Fig. 7), LLQ and PQ algorithms managed to squeeze those values below 100 ms, while the CQ has under 255ms, which already exceeds the threshold set by the QoS requirement.It is clear that in the case of WFQ and CBWFQ the delay is a little more than 1 second, while in the case of FIFO than can reach up to 8.5 seconds, which are unacceptable values provided by the QoS requirements.As for the delay variation (jitter) (see Fig. 8 and Fig. 9) the LLQ, PQ and CQ has managed to keep the measured values below 30ms.Subsequently, the WFQ and CBWFQ ensured around 150ms and 210 ms jitter, while the FIFO has finally managed to stabilize its delay variation around 1 second.It www.ijacsa.thesai.orgshould be noted that in terms of jitter PQ and LLQ congestion management algorithms managed to meet under the requirements of the QoS threshold requirement.

IV. CONCLUSION
This paper compares the performance of the main congestion management algorithms based on a measurement environment.The laboratory environment was implemented at the Faculty of Informatics University of Debrecen.In all cases the measurement result shows that the FIFO scheduling principle is the most inconvenient algorithm for handling of interactive voice packets in case of congestion.In the case of voice transmission the PQ and the LLQ algorithms were the two most appropriate algorithms, in terms of packet loss rate, end-to-end delay and jitter.Using these algorithms no packets suffered packet loss.However, knowing the principle of the PQ, namely that it serves the maximum priority queue, but produce packet starvation for other tree queues, based on the literature and on the measurement results the conclusion is, that for the interactive real-time voice traffic, taking all in consideration, the LLQ congestion management algorithm is most appropriate.Further research topic is to support the results and test the algorithms presented in the current article by mathematical modeling.

Fig. 4 .
Fig. 4. VoIP packet lossFig.5shows the same content as Fig.4, except that the former does not include the efficiency representation of the two least efficient congestion management algorithm.Thus it is prominently observable the difference of performance of PQ, WFQ, CBWFQ and LLQ in packet loss.Easy to see, that in the case of LLQ and PQ there was no packet loss due to the absolute priority queue, in which the real-time voice was

PSTN WAN Main Campus Remote Campus Campus Backbone Campus Backbone Server Farm
Fig. 1.A classical non-converged network