Você está na página 1de 5

Journal of Informatics & Electronics, Vol.2, No.2, pp.

15-19, March 2008

Adaptive Weighted Fair Queueing with Pritority (AWFQP) Scheduler for Diffserv Networks
I-Shyan Hwang, 1Bor-Jiunn Hwang, and Ching-Shyan Ding
Department of Computer Engineering and Science, Yuan-Ze University ishwang@saturn.yzu.edu.tw
1

Department of Computer and Communication Engineering, Ming Chuan University bjhwang@mcu.edu.tw Department of Computer Engineering and Science, Yuan-Ze University ding.william@delta.com.tw Abstract

DiffServ networks support three forwarding classes: EF (Expedited Forwarding), AF (Assured Forwarding) and BE (Best Effort). Scheduling algorithms based on priority queueing, such as Priority Queueing with Weighted Round-Robin (PQWRR) are the traditional choices for providing EF service. The Dynamic DiffServ Scheduling (DDS) scheduler can improve the starvation problem for the BE and AF traffics, but scarify the EF performance. This paper proposes an Adaptive Weighted Fair Queueing with Priority (AWFQP) scheduler to provide QoS (Quality of Service) guarantees to EF, AF and BE classes. The objective of the proposed scheduler is to improve the performance of AF and BE classes without scarifying EF performance. Simulation results show that the proposed AWFQP scheduler outperforms the PQWRR and DDS schedulers in guaranteeing QoS (Quality of Service) for the AF and BE classes, improves system throughput and reduces jitter and delay time. Keywords: PQWRR, QoS, AWFQP, DDS.

1. INTRODUCTION
The Internet provides only one service, namely, best effort service to all applications. Under the best effort approach, all packets are treated the same in a first-come-first-served basis and no scheme exists for differentiating the services for specific applications. Conventional Internet services effectively support non-real-time applications such as HTTP or FTP. However, with the increasing demands of supporting the multimedia real-time traffic, such as Internet telephony, video conferencing, interactive gaming and mission-critical business traffic, reliable services with QoS (Quality of Service) guarantee are required. The DiffServ architecture [1] sends complicated functionality to the edge routers (ER) of the DiffServ domain and keeps core routers (CR) simple. Core routers of the domain do not need to maintain per-flow state information, instead must forward packets according to the per hop behavior (PHB) associated with each traffic class. ERs maintain all user traffic profiles; that is, they classify, meter, and shape all incoming packets to ensure that

individual user traffic flow conforms to the service level agreement (SLA) specified by a network operator. An SLA is the service contract a customer (can be an organization or an upstream service provider) signed with its service provider to guarantee certain PHBs. Traffic complying with the SLA is allowed to enter the network regardless of destination. The DiffServ network architecture is shown in Figure 1.1. The DiffServ architecture standardizes a set of PHBs that include Expedited Forwarding (EF) [2], Assured Forwarding (AF) [3] and Best Effort (BE). Expedited
D iffServ D om ain S O ther D iffServ D om ain CR SLA SLA CR BR

C lient site R ER ER

BR

SLA : 1 M bps of EF 4 M bps A F 5 M bps B E

ER : Edge R outer B R : B order R outer C R : C ore R outer

Figure 1.1 The DiffServ network architecture

- 15 -

Adaptive Weighted Fair Queueing with Pritority (AWFQP) Scheduler for Diffserv Networks Forwarding (EF) of PHBs is the primary technique in DiffServ for providing a low-loss, low-latency, low-jitter, and assured bandwidth service. EF can be implemented by utilizing priority queueing with rate limiting on the class. Real-time applications with stringent delay requirements, such as VoIP and interactive gaming, are particularly suited to be forwarding utilizing EF. Assured Forwarding (AF) of PHBs is defined to provide different forwarding assurances, and the AF PHB defines four AFx classes: AF1, AF2, AF3, and AF4. Each class is assigned a specific amount of buffer space and interface bandwidth to guarantee a precise QoS [4]. Within each AFx class, three-drop precedence values are specified. Under congestion, packets marked with high drop precedence are dropped first. Therefore, packets within the same AFx class will experience similar QoS in delay and jitter and different QoS in loss rate. In general, packets are marked in accordance with service agreements with the service provider. Packets exceeding this service profile are marked with high drop precedence and dropped first when congestion occurs. Non-real-time applications, such as streaming video, can use the AF service. Best Effort (BE) PHB provides no service guarantee. All packets belonging to BE class is treated the same. Best effort service can no longer meets diverse QoS requirements. All IP packets in the BE class are not policed and are forwarded at best effort. Implementation of PHBs relies considerably on scheduling and queueing techniques in switches and routers. Several proposals supporting scheduling schemes are available for output queueing switches, such as priority queueing with weighted round- robin (PQWRR) [5][6] and for input queueing switches, such as dynamic DiffServ scheduling (DDS) algorithm [7]. The PQWRR algorithm combines the PQ and WRR scheduling, and assigns EF traffic a higher priority than AF and BE traffic and, therefore, reduces EF traffic delay and jitter. However, the starvation is occurred for the BE traffic. The DDS algorithm is to schedule EF and AF traffic according to their minimum service rates with the reserved bandwidth and schedule AF and BE traffic fairly with the excess bandwidth. The problem of starvation was overcome in the DDS; nevertheless the bandwidth utilization is not studied. In this paper, simulation results indicate that the proposed AWFQP scheduler has the best performance than those of PQWRR and DDS, especially for the traffic load is heavy, and provides QoS bandwidth guarantees for EF and AF traffic and avoids the starvation for BE traffic when traffic load is heavy.

2. THE PROPOSED AWFQP SCHEDULER


The AWFQP scheduler, shown in Figure 2.1, is a hierarchical scheduler. In the first level, it is constructed based on the Priority Queueing Scheduler and Fair Queueing Scheduler for DiffServ network. The Priority Queueing Scheduler of the AWFQP, EF packets always has the highest priority to use the allocated bandwidth, and BE packets get served only if there are no EF packets waiting. The purpose of Priority Queueing Scheduler can improves the starvation problem for BE traffic, but without scarify EF performance. Each AF class has its own bandwidth reservation in Fair Queueing Scheduler of the AWFQP, so that AF class performance and QoS guarantee can be improved. In the second level, the Adaptive Queueing Scheduler is based on a traffic threshold (TH). When traffic load (TL) is lower than the traffic threshold, bandwidth provisioned allocated to the EF class can be shared by the BE class based on the Priority Queueing Scheduler. The Adaptive Queueing Scheduler is operated as follows: if the result of Priority Queueing Scheduler is EF class, then Adaptive Queueing Scheduler is changed to the Priority Queueing Scheduler, on the other hand, if the result of Priority Queueing Scheduler is BE class, then adaptive queueing scheduler is changed to the Fair Queuing Scheduler; otherwise, when traffic load exceeds the traffic threshold, BE class shares bandwidth with AF class based
The algorithm of the adaptive queueing scheduler

Adaptive queueing schedule (TH,TL) { TH, = Traffic threshold; TL= Traffic loads; BEGIN If TLTH { BE class can use bandwidth of EF class; If output = EF Adaptive queueing schedulerPriority Queueing Scheduler Else Adaptive queueing schedulerFair Queueing Scheduler } Else { BE class share bandwidth with AF class; Adaptive queueing schedulerPriority Queueing Scheduler } END }

- 16 -

Journal of Informatics & Electronics, Vol.2, No.2, pp.15-19, March 2008

3.2 PERFORMANCE EVALUATION


F irst lev el
P rio rity Q u e u in g S ch e d u ler

S eco nd lev el
EF Throughput

EF

Y es
TL

BE AF1 AF2 AF3 AF4

TH

No F air Q u e u in g S ch e d u ler

A d a p tiv e Q u eu in g S ch ed u le r

5400 4900 4400 3900 3400 2900 2400 1900 0 20 40 60 Traffic Load (%) 80 100

Kbps

(a) EF class Figure 2.1 Adaptive Weighted Fair Queueing scheduler with Priority (AWFQP) scheduler.
2400 AF Throughput

on the Fair Queuing Scheduler, then the adaptive queueing scheduler is changed to the Priority Queueing Scheduler, the operation of the scheduler is the same as that for PQWRR.

2000 Kbps 1600 1200 800 400 0 0 20 40 60 Traffic Load (%) 80 100

3. SIMULATION RESULTS
3.1 SIMULATION ENVIRONMENT
1200

(b) AF class
BE throughput

This section compares system performance of the AWFQP algorithm with the PQWRR algorithm and DDS algorithm in terms of throughput, delay and jitter. The system parameters for simulation are shown in Table 3.1 [1], and the traffic arrive rates are Poison distribution with different mean values 100 (EF), 100 (AF) and 50 (BE), respectively. The initial traffic loading for EF, AF and BE traffic for AWFQP are with different values 1.9M, 2.5M, and 1.1M, respectively. The QoS requirements of each DiffServ classes are different; for EF class, low delay service is expected; for AF classes, the delay requirements are not as stringent as EF class but the delays should still be bounded; for BE class, service starvation should be avoided by guaranteeing a minimum throughput.
Table 3.1 System parameters for simulation

1000 Kbps 800 600 400 200 0 0 20 40 60 Traffic Load (%) 80 100

(c) BE class

Class 1 Data Type Priority Packet interval Data Rate Data Size Minimum bandwidth

Class 2

Class 3 BE (Data) Low 20 ms

EF (Audio) AF (Video) High 10 ms Medium 10 ms

Figure 3.1 Throughput comparison of AWFQP between PQWRR and DDS with different traffic thresholds 25%, 50%, 75% for (a) EF (b) AF (c) BE when increasing the EF class traffic load.

36.8 Kbps 1171.2 Kbps 1200 Kbps 92 Bytes 512 Kbps 1464 Bytes 512 Kbps 1500 Bytes NC

Figure 3.1(a) shows a comparison of throughput of EF for AWFQP, PQWRR and DDS with different thresholds of 25%, 50% and 75%, respectively. The proposed algorithm performs better than the DDS and close to the PQWRR when the traffic load is below 80%. Figures 3.1(b) and Figure 3.1(c) present the throughput condition for the BE and AF traffic classes for AWFQP, PQWRR and DDS in different thresholds of 25%, 50%, and 75%, respectively. - 17 -

Adaptive Weighted Fair Queueing with Pritority (AWFQP) Scheduler for Diffserv Networks The proposed algorithm (AWFQP) outperforms PQWRR and DDS when the traffic load is high and avoids the BE traffic starvation. The possible reason is that less bandwidth is occupied by BE in PQWRR and DDS schedulers in higher traffic loading situation.
35 30 25 ms 20 15 10 5 0 20 40 60 Traffic Load (%) 80 100 AWFQP 75% PQWRR DDS AWFQP 50% AWFQP 25%

Figure 3.3 Jitter distribution comparison of EF class for AWFQP and PQWRR scheduler with traffic thresholds 25%, 50% and 75% when increasing the EF class traffic load.

(a) EF class

(b) AF class

(c) BE class

Figure 3.2 shows the delay comparison for the EF, AF and BE traffic between the AWFQP, PQWRR and DDS with a threshold of 75%. The trend in Fig 3.2 is similar to that in Fig 3.1. The EF performance of the AWFQP is better than DDS and almost closes to the PQWRR. However, the AWFQP outperforms the DDS and PQWRR in AF and BE traffic. As predicted, AWFQP are guaranteed to meet the stringent delay requirement by allocating more bandwidth. Compared to the PQWRR and DDS scheduler, the AWFQP provides service isolation and hard QoS to the AF and BE classes. It means that the AWFQP is suitable for multimedia transmission of DiffServ network, and AWFQP has better bandwidth utilization especially for high traffic loading case. Figure 3.3 shows the jitter distribution of EF traffic for PQWRR, DDS and AWFQP with different traffic thresholds when increasing the EF class traffic load. The figure shows that the AWFQP has the best performance than those of PQWRR and DDS, especially for the heavy traffic load due to more even transmission opportunity for EF traffic. The system performance comparison is summarized in Table 3.2. The simulation results indicate that the best performance is when the threshold is 75% which is an excellent reference can be used by bandwidth provisioned allocated in the AWFQP, PQWRR and DDS schedulers.
Table 3.2 Performance comparison between AWFQP, PQWRR and DDS for EF, AF, and BE.

PQWRR EF AF BE
Figure 3.2 Delay comparison of AWFQP between PQWRR and DDS with traffic threshold 75% for (a) EF (b) AF (c) BE class when increasing the EF class traffic load.

DDS Worst Close to PQWRR Better

AWFQP Close to PQWRR Best Best

Best Worst Worst

- 18 -

Journal of Informatics & Electronics, Vol.2, No.2, pp.15-19, March 2008

4. CONCLUSION
This paper proposed an AWFQP scheduler based on traffic thresholds (TH). Simulation results showed that the proposed AWFQP algorithm provides better EF class performance than the DDS and close to the PQWRR, and outperforms the PQWRR and DDS for BE and AF classes performance. The proposed AWFQP scheduler can guarantee the QoS for the EF and AF traffics and avoid the starvation for BE traffic when the traffic load is high. Overall, the AWFQP can improve the resource utilization while providing end-to-end QoS guarantees to all DiffServ classes.

[2] [3] [4]

[5]

[6]

[7]

V. Jacobson, K. Nichols and K. Poduri (1999). An Expedited Forwarding PHB. IETF RFC 259. J. Heinanen, F. Baker, W. Weiss and J. Wroclawski (1999). Assured Forwarding PHB Group. IETF RFC 2597. N. Michael and C. Tiglao (2007, 3). Utility-based enhanced priority scheduler for Differentiated services. in Proc. IEEE AMS, 187-192. J. Mao, W.M. Moh and B. Wei (2001). PQWRR scheduling algorithm in supporting of DiffServ. in Proc. ICC 2001, 3, 679-684. Y. Zhang and P.G. Harrison (2007, 8). Performance of a priority-weighted round robin mechanism for Differentiated service networks. in Proc. IEEE ICCCN, 1198-1203. M. Yang, E. Lu and S.Q. Zheng (2003). Scheduling with dynamic bandwidth allocation for DiffServ classes. in Proc. ICCCN 2003, 319-324.

REFERENCES
[1] I.S. Hwang, B.J. Hwang and Y.S. Liu (2003, 3). QoS guarantee in interconnected DiffServ based on MPLS network. Journal of Technology, 18(2), 153-164.

Received 1 Reviewed 2nd Reviewed Accepted


st

2007/10/16 2008/02/05 2008/03/11 2008/04/08

- 19 -

Você também pode gostar