In this paper we consider the use of a supercomputer with a hardware shared memory versus a cluster of workstations using a software Distributed Shared Mem-ory (DSM). We focus on ray tracing applications to compare both architectures. We have ported Stingray, a parallel cone tracer developed on a SGI Origin 2000 super-computer, on a cluster using a Scalable Coherent Interface (SCI) network and a software DSM called SciFS. We present concepts of cone tracing with Stingray, concepts of SCI cluster with a DSM and the implementa-tion issues. We compare the results obtained with the two architectures and we discuss the trade-off - price/performance/programming ease - of both architectures. We show with Stingray that a modest 12 nodes SCI cluster with an efficient software DSM is 5 times cheaper and can perform up to 2.3 times better than a SGI Origin 2000 with 6 processors. We think that a software DSM is well suited for this kind of applications and provides both ease of programming and scalable per-formance.

@inproceedings{MC01,
author = {Alexandre Meyer and Emmanuel Cecchet},
title = {Stingray: Cone Tracing using a software {DSM} for {SCI} clusters},
booktitle = {3rd IEEE International Conference on Cluster Computing, CLUSTER'01, October, 2001},
address = {Newport Beach, CA, Etats-Unis},
publisher = {IEEE},
pages = {95--101},
year = 2001
}