Performance evaluation of forwarding algorithms for generalized storage aware routing protocols

This thesis presents an investigation of the design and evaluation of the generalized storage aware routing (GSTAR) protocol proposed for use in the MobilityFirst future Internet architecture. The GSTAR protocol uses in-network storage to improve service quality and throughput in wireless access networks with varying radio link quality and/or disconnection. These gains are achieved using a combination of short-term buffering at routers to smooth out fluctuations in path quality along with delay-tolerant storage, to overcome total disconnection of the mobile device. The performance of the GSTAR protocol is evaluated for exemplary wireless access network scenarios using ns-3 based simulation models, and key design parameters are investigated. Each node in GSTAR maintains two kinds of topology information. The intra-partition graph contains information about path quality between nodes in the current partition of the network. The path quality is determined using two metrics: short term and long term expected transmission time (SETT and LETT). Every node compares these two metrics using the store/forward decision threshold and stores the data on finding that the path is degraded with the expectation that it may improve in the future. Inter-partition graph gives a probabilistic view of the connection patterns between nodes in the network. It is used in the event of disconnections or partitions. An ns-3 based simulation model is described which includes nodes with storage, hop-by-hop transport, time-varying wireless channels and mobile users with possible disconnection. The model is used to evaluate different forwarding algorithms in GSTAR. Using a baseline threshold scheme where packets are temporarily stored when SETT > 1.1 * LETT, it is shown that the resulting system achieves performance improvements over the baseline with no storage. The threshold algorithm is studied further to consider adaptive settings based on the moving average and other temporal filters of the SETT sequence. The results show that if link quality fluctuations are random, the moving average scheme works well, while an exponentially weighted moving average is recommended for on-off channels with periodic outages. Simulation results are provided in each case, showing the benefit of adaptive threshold settings over the baseline non-adaptive case considered in earlier work.M.S.Includes bibliographical referencesby Nehal Soman