A New Approach to Rate-Based Traffic Control Methods

1-The source queue sizes are infinite, while the output buffer of the switch has a finite capacity. Source queue nodes have the same control parameters (token pool size, etc.). In feedback-based rate control model, in case of high occupancy of multiplexer buffer, source node emission rates are reduced with the same ratio, and at the same time, which is determined by addition of a given amount of delay with the current time. 2-Both periods of burst and silence are assumed to be exponentially distributed with constant average lengths during the simulation. 3-The interarrival period of the cells during the burst mode are also assumed to be exponentially distributed. The average arrival rate which is the same as the peak cell rate, are reduced to decrease the average cell arrival rate. 4- Service rates are assumed to have a fixed ratio with respect to the average arrival rates. Here the ratio is set equal 1.11 which gives a 90 percent utilisation. 5-Multiplexing gain which is defined as the peak to the allocated service rate is assumed to be the same during the experiments, 6-Traffic source nodes are assumed to be of the same type. In these simulations, in order to keep burstiness constant, we decided to change the peak arrival rate as a means of changing the incoming traffic to the source and multiplexer buffers. Burstiness is defined as in Eq(8): .
mmp
(8);
where ut is utilisation, and avgsil and avgburst are respectively the average silence and average burst duration in average cell interarrival units, and k is the burstiness factor. Here as a result of constant ratio between service and arrival rate, ut is constant. Fig5 to Fig8 show that, for both feedback based-rate control and leaky bucket policing method, reducing peak-cell-rate, with a constant ratio to the source and multiplexer buffer service rates, increases delay and delay variance for both control type. But the effect is more noticeable for leaky bucket control compared to feedback-based rate control method. Intuitively delay increase for leaky bucket control can be attributed to the lower service rate of each source queue as well as to multiplexer allocated capacity of the outgoing link.
Figure 5 Delay versus average arrival rate for feedback-based rate control
Figurerarrival for leaky bucket control
k = ( ut ) ( avgsil ) = ( 1 – ut ) ( avgburst )
Figure 3: Delay versus service-to-average arrival rate ratio ( arrival rate =0.01) 4. SIMULATION MODELS AND RESULTS The simulation model has been implemented using OPNET simulation package. The source and switch/ multiplexer queues are shown in Fig3. In this figure ‘mmp’ nodes are interrupted poisson process traffic source types, Sq1 to Sq10 are the source module queues, SW is the lan or workstation local ATM switch module, and Sink node is the process model used for calculation and storing of the obtained results. So performance is calculated for the combination of a two stage queuing system. The assumptions made for the study of the control schemes are described as follows: Feedback Control Loop Sq1 mmp Sq2 : : : mmp Sq10 Figure 4: Simulation model for the study of rate control schemes. . : : : Sink node
Figure 8 Delay variance versus average interarrival for feedback-based rate control 5. CONCLUSION The results obtained from the above experiments, has lead us to specify the effect of constant service-toaverage arrival rate, on the performance measures of preventive and closed-loop rate control mechanisms. This study and results have been carried out using both analytical and simulation models. The obtained results show that in obtaining the same QOS measures such as delay and delay variance for leaky bucket or feedbackbased rate control schemes, a higher ratio of service-toarrival rate is necessary for low average rate sources. This study has also shown that a constant ratio of service-to-average arrival rate has comparatively less effect on feedback-based rate control performance. Finally an accurate assessment of delay and delayvariance, requires the average load of the traffic source to be considered, in parallel to other traffic factors such as burstiness. REFERENCES Figure 6 Delay versus average interarrival for leaky bucket control [1] H. Ahmadi, R. Guerin and K. Sohraby, "Analysis of Leaky Bucket Access Control Mechanism with Batch Arrival Process", Proc. IEEE GLOBECOM ’90 (San Diego, December 1990), pp. 344-349. [2] K. Sohraby, M. Sidi, "On the Performance of Bursty and Correlated Sources Subject to Leaky Bucket Rate-Based Access Control Schemes", Proc. IEEE INFOCOM ’91, (Bal Harbour, April 1991), pp.426-434. [3] S. Wittevrongel, H. Bruneel, "Output Traffic Analysis of a Leaky Bucket Traffic Shaper Fed by a Burst Source" Proc. of ICC’94, (), pp.1581-1585
This means that if a cell enters a source queue with empty token pool, it has to be delayed for a duration approximately equal to the product of token generation period and source queue length. In feedback-based rate control scheme, delay increase is simply due to a lower allocated service rate. As a result for the same queue size, delay will be greater for lower service rate of cells. From the obtained results we conclude that, in achieving a given QOS level, a higher service-toarrival ratio should be allocated to lower average rate of traffic, and that this ratio depends on the traffic control scheme.