Clusters composed of fast personal computers are becoming more and more attractive as cheap and efficient platforms for distributed and parallel applications. The main drawback of a standard NOW (Network Of Workstations) is the poor performance of the standard inter-process communication mechanisms based on RPC, sockets, TCP/IP, Ethernet. Such standard communication mechanisms perform poorly both in terms of throughput and of message latency.
Recently, several prototypes developed around the world have proved that re-visiting the implementation of the communication layer of a standard Operating System kernel, a low cost hardware platform composed of only commodity components can scale up to several tens of processing nodes and deliver communication and computation performance exceeding the one delivered by the conventional high-cost parallel platforms.
Despite the importance of this break-through, that allows the use of inexpensive hardware platforms for efficient support of large/medium/fine grain parallel computation in a NOW environment, few papers describing their design and implementation still appear in the literature. Multiprogramming and co-scheduling of communicating processes so that "real applications" can be efficiently parallelized in NOW environments are still open research issues as well.
This workshop will provide a forum where researchers and practitioners can discuss issues, results, and ideas related to the design of efficient NOWs based on commodity hardware and public domain operating systems as compared to custom hardware devices and/or proprietary operating systems.
email@example.com, June 10, 1998