The MuSE Runtime System for SCI Clusters: A Flexible Combination of On-Stack Execution and Work Stealing

* Final gross prices may vary according to local VAT.

Get Access

Abstract

Beyond its high bandwidth and low latency properties, the Scalable Coherent Interface (SCI) technology offers capabilities for a shared-memory communication paradigm on distributed systems. In particular, networked workstations and PCs can basically be transformed into NUMA machines. As such, alternative execution models become possible that were otherwise infeasible on networks of workstations. With LANs, these normally rely on an inadequate communication infrastructure and they are fixed to the message-passing paradigm.