No abstract

Network based congestion avoidance which involves managing the queues in the network devices is an integral part of any network. Most of the mobile networks today use Droptail queue management where packets are dropped on queue overflow. Droptail, however, is known to suffer from the well known global synchronisation problem which is characterised by the phenomenon of alternating periods of empty and full queues and hence bursty losses. Especially in resource constrained networks such as MANETs, packet loss results in increased overhead in terms of energy wasted to forward a packet which was eventually dropped, additional energy required to retransmit this packet and the degraded service quality as experienced by the end user application. Active Queue Management (AQM) has been successfully demonstrated as a solution to the global synchronisation problem in the context of wired networks. However, if AQM is to be deployed in MANETs, it should be lightweight, proactive and easy to implement as mobile networks are resource constrained in terms of memory, processing power and battery life. To the best of our knowledge a study addressing the implications of AQM in mobile networks (MANETs in particular) does not exist. This paper presents a predictive queue management strategy named PAQMAN that proactively manages the queue, is simple to implement and requires negligible computational overhead (and hence uses the limited resources efficiently). The performance of PAQMAN (coupled with Explicit Congestion Notification -ECN)has been compared with Droptail through ns2 simulations. Results from this study show that PAQMAN reduces packet loss ratio (and hence the fraction of retransmissions) while at the same time increasing transmission efficiency. Moreover, as its computational overhead is negligible, it is ideally suited for deployment in MANETs.

Several Active Queue Management (AQM) based solutions have been proposed to enable service differentiation in the DiffServ Assured Forwarding (AF) class(es). Most of these solutions, however, provide throughput guarantees only. This paper proposes a new queue management approach called PAQMAN-DS which provides quantitative controlled delay guarantees to delay sensitive applications in the AF class. The proposed approach is based on predicting the future state of the queue and requires specification of only a single parameter (target delay) per hop. Performance evaluation of PAQMAN-DS through ns-2 simulations reveals that it regulates the delay around the target mark on a per-hop basis, discriminates in favour of IN contract traffic and simultaneously achieves high link utilization.

Due to increasing reliance on computer communication networks, it is highly desirable that networks should have the ability to detect symptoms of oncoming exception conditions and take measures to prevent them thereby enabling a degree of Proactive Network Management that underpins an acceptable Quality of Service. This paper proposes a framework for achieving congestion avoidance through Proactive Network Management using data mining. It examines the inter-relationships between network element Management Information Base (MIB) attributes, queue parameters (associated with a transmission link) and the level of congestion at a network node and identifies hybrid parameters that have a bearing on congestion. By employing data mining on the data pertaining to these variables, congestion at the network node can be predicted. Results from our initial experimentation with particular data mining models show that the accuracy achieved is as high as 98% in all of the cases thus rendering data mining a viable approach to proactively identify network exception conditions.

No abstract