Scalable Shared Memory and Shared Memory Clustering

Abstract

Users want tightly coupled, shared everything systems that are highly available, scalable, have a single system image, and are cheap. Unfortunately for computer customers (and fortunately for computer engineering research), these goals conflict. The talk will discuss possible design points and trade-offs that can be pursued in an attempt to reconcile these conflicting goals.

We shall survey ongoing research on a scalable shared memory architecture at the IBM T.J. Watson Research Center which explores one such interesting design point, namely shared I/O and shared memory support in an integrated, scalable, distributed OS SP-like system.