DSM perspectives: Another point of view

Distributed shared memory computers (DSM's) have arrived [4] . [5] to challenge mainframes. DSM's scale to 128 processors with two to eight processor nodes. As shared emory multiprocessors (SMP's). DSM's provide a single system image and maintain a "shared everything" model. Large-scale UNIX servers using the SMP architecture challenge mainframes in legacy use and applications. These have up to 64 processors and a more uniform memory access. In contrast, clusters both complement and compete with SMP's and DSM's, using a "shared nothing" model. Clusters built from commodity computers, switches, and operating systems scale to almost arbitrary sizes at lower cost while trading off SMP's single system image. Clusters are required for high availability applications. Highest performance scientific computers use the cluster (or MMP1) approach. High growth markets, e.g., Internet servers, online transmission processing (OLTP), and database systems can all use clusters. The mainline future of DSM may be questionable because: small SMP's are not as cost effective unless built from commodity components; large SMP's can be built without the DSM approach: and clusters are a cost-effective alternative for most applications to SMP's, including DSM's, for a wide scaling range. Nevertheless, commercial DSM's are being introduced that compete with SMP's over a broad range.