SCALABILITY STUDY OF THE KSR-1

Scalability of parallel architectures is an interesting area of curren t research. Shared memory parallel programming is attractive stemming from its relative ease in transitioning from sequential programming. H owever, there has been concern in the architectural community regardin g the scalability of shared memory parallel architectures owing to the potential for large latencies for remote memory accesses. KSR-1 is a commercial shared memory parallel architecture, and the scalability of KSR-1 is the focus of this research. The study is conducted using a r ange of experiments spanning latency measurements, synchronization, an d analysis of parallel algorithms for three computational kernels and an application. The key conclusions from this study are as follows: Th e communication network of KSR-1, a pipelined unidirectional ring, is fairly resilient in supporting simultaneous remote memory accesses fro m several processors. The multiple communication paths realized throug h this pipelining help in the efficient implementation of tournament-s tyle barrier synchronization algorithms. Parallel algorithms that have fairly regular and contiguous data access patterns scale well on this architecture. The architectural features of KSR-1 such as the poststo re and prefetch are useful for boosting the performance of parallel ap plications. The sizes of the caches available at each node may be too small for efficiently implementing large data structures. The network does saturate when there are simultaneous remote memory accesses from a fully populated (32 node) ring.