miNI: reducing network interface memory requirements with dynamic handle lookup
ICS '03 Proceedings of the 17th annual international conference on Supercomputing
Hi-index | 0.00 |
Clusters of high-end workstations are currently used in many application domains to perform large-scale computations or as scalable servers for I/O bound tasks. Although clusters have many advantages, their applicability in emerging areas has been limited. One of the reasons for this is the fact that clusters do not provide a single system image. In this work we address this by providing a single cluster image for thread and memory management. We implement our system, CableS, on a 32-processor cluster. We demonstrate the versatility of CableS with a range of applications. We run pthreads applications. We use an OpenMP-to-pthreads compiler on OpenMP programs and execute them. We provide an implementation of M4 macros and run SPLASH-2 applications. Fianlly, we show that the overhead introduced by CableS affects the parallel section of applications only in cases where the data placement is affected by operating system limitations in virtual memory mappings granularity.