Group Scalability in Distributed Systems
We address shortcomings of two important group communication paradigms, IP Multicast and gossip based message dissemination, both of which have scalability issues when the number of groups grows.
First, we propose a transparent and backward-compatible layer called Dr. Multicast to allow data center administrators to enable IPMC for large numbers of groups without causing scalability issues. Dr.Multicast optimizes IPMC resources by grouping together similar groups in terms of membership to minimize redundant transmissions as well as the cost of filtering unwanted messages.
Second, we argue that when nodes belong to multiple groups, gossip based communication loses its appealing property of using a fixed amount of bandwidth. We propose a platform called GO (Gossip Objects) that bounds each node’s bandwidth use to a customizable limit, prohibiting applications from joining groups that would cause the limit to be exceeded. We make the observation that gossip rumors tend to be small, and propose a utility-based heuristic to stack rumors into packets to optimize delivery speed, with rumors sometimes travelling through indirect paths.
Features joint work with Hussam Abu-Libdeh, Mahesh Balakrishnan, Ken Birman, Robert Burgess, Gregory Chockler, Qi Huang, Jure Leskovec, Haoyuan Li, Deepak Nataraj and Yoav Tock.
Seminar Ymir VIGFUSSON (IBM Research, Haifa) – Thursday 19 November, 2pm
INRIA Rennes – Bretagne Atlantique, Salle Sardaigne