|M.Sc Student||Shoham Moshe|
|Subject||SSM+: Enhanced Algorithm for Scalable Session Message in SRM|
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus Adrian Segall|
SRM (Scalable Reliable Multicast) is a reliable multicast protocol aimed to be scalable and robust under a wide range of group sizes, dynamic group membership and dynamic topologies. SRM members exchange periodically session messages in order to detect data loss and to compute the delay to other members. SSM (scalable session messages) was designed to overcome the scalability problems of the SRM session messages. SSM uses a stochastic distributed self-configuration algorithm in order to determine a hierarchical structure of global and local members. Global members are used as representatives and only they can transmit global session messages to the entire multicast group. SSM+ is suggested in this work to improve SSM and to enhance its ideas. The work indicates and solves issues that cause sub-optimal performance in SSM including a dynamic hierarchy configuration failure scenario. SSM+ limits the randomness of SSM in order to better guide it towards more desirable results. This is achieved by imposing constraints that limit the distance of a local member to its representative, by enhancing the SSM appropriateness measures to include immunity to state changes and by introducing the relative proximity idea. SSM+ determines differently the local scope of local members messages and revises the distance estimation mechanism to produce more accurate delay estimations. The work also investigates the use of a new suggested local opportunism mechanism that enables a local node to switch a representative when it learns of the existence of a closer one. Simulation results validate the intuitive arguments that SSM+ is more robust to group dynamic changes than SSM and improves its delay estimation accuracy, while maintaining or reducing the SSM bandwidth consumption.