A comprehensive bibliography of distributed shared memory
ACM SIGOPS Operating Systems Review
IEEE Parallel & Distributed Technology: Systems & Technology
| Hi-index | 0.00 |
Abstract Parallel computing on a distributed system, such as a network of workstations, can saturate the communication network, leading to excessive message delays and consequently poor application performance. Current operating systems offer only partial support for flow control protocols that can help insulate application performance from extraneous traffic on the shared network. We examine empirically the consequences of integrating one such protocol, called Warp control~\cite{Park93}, into Mermera, a software shared memory system that supports parallel computing on distributed systems~\cite{HeddayaS93hicss}. Preliminary performance measurements are reported for an asynchronous iterative program to solve a system of linear equations, under varying levels of network contention. The experiments were conducted on a network of seven Sun Sparc~1+ workstations, using an auxiliary traffic generator. These measurements show that Warp succeeds in stabilizing the network behavior when there is high contention, increasing the effective throughput available to the application, and consequently decreasing its completion time. In some cases, however, Warp control does not achieve the performance attainable by fixed size buffering when using a statically optimal buffer size. Based on the nature of Warp and the underlying communication layers, we offer explanations for our results. Our use of Warp to regulate the allocation of network bandwidth emphasizes the possibility for integrating it with the allocation of other resources, such as CPU cycles and disk bandwidth, so as to optimize overall system throughtput, and enable fully-shared execution of parallel programs.