Chandana Watagodakumbura

Aggregated priority queueing as an efficient queueing mechanism to overcome effects of self-similarity

Differentiated Services (DiffServ) architecture is based on aggregation of traffic and, unlike the Integrated Services (lntServ) the resources are not reserved on a per flow basis. One of the current research issues is the provision of Quality of Service (QoS) guarantees to real time traffic in the DiffServ environment. The idea here is to achieve these guarantees without per flow resource allocation while safeguarding the best effort traffic from extremely high delays. The self-similar nature of Internet traffic that results in burstiness and very high delays has been identified in the recent past. The problem at hand is, what packet-scheduling scheme can be used to minimise delay to reasonable limits and the levels of control to be applied on the aggregated traffic to achieve such limits. We propose a class based aggregated Priority Queue with Lower Real time Traffic Utilisation (PQ-LRTU) as a packet-scheduling scheme that is capable of providing statistical delay guarantees for DiffServ networks. Since only a real time traffic utilization threshold is maintained at the entrance to the network to control traffic, it provides a simple mechanism. Overall queue efficiency is achieved by using different levels of delay tolerable lower priority traffic.

Effects of traffic aggregation on quality of service parameters

Differentiated Services (DiffServ) Architecture is based on aggregation of traffic as opposed to per flow traffic. When DiffServ is used in real time traffic environment it is important to study the effects of traffic aggregation on quality of service parameters. One-way delay and instantaneous packet delay variation (IPDV) are the most commonly used quality of service parameters for real time traffic. In this paper we study how one-way delay and IPDV are affected by traffic aggregation. We discover through our simulations and analytical models that IPDV increases with traffic aggregations while one-way delay is not affected purely due to aggregation. We used a priority queue scheduler for our simulation where the real time traffic was served with highest priority.

Statistical multiplexing gains in the presence of aggregated self-similar traffic explained using binomial distribution

Differentiated Services (DiffServ) Architecture is based on aggregation of traffic as opposed to per flow traffic. When DiffServ is used in a real time traffic environment, it is important to Study the effects of traffic aggregation on quality of service parameters. One-way delay and instantaneous packet delay variation (IPDV) are the most commonly used quality of service parameters for real time traffic. Also, fractal or self-similar nature of the Internet traffic has been identified in recent years. In this paper, we study how one-way delay and IPDV are affected by aggregation of self-similar traffic. We make quantitative comparisons of IPDV and one-way delay between aggregated self-similar traffic and exponential inter-arrival traffic, when the only traffic control at the entrance of the network is a real time traffic utilization threshold. Under these conditions, we point out that considerable statistical multiplexing gains are possible with respect to delay. Further, we bring out the notion of ¿level of active sources that makes a significant impact on delay¿, based on binomial distribution, to explain above statistical multiplexing gains. We also bring out that such multiplexing gains are not possible with IPDV as it is based on each flow, unlike on the aggregate as in the case of delay.