Lazy direct-to-cache transfer during receive operations in a message passing environment
Proceedings of the 3rd conference on Computing frontiers
MEDEA '05 Proceedings of the 2005 workshop on MEmory performance: DEaling with Applications , systems and architecture
Microprocessors & Microsystems
ISPA'07 Proceedings of the 5th international conference on Parallel and Distributed Processing and Applications
| Hi-index | 0.00 |
With increasing uniprocessor and SMP computation power, workstation clusters are becoming viable alternatives to high performance computing systems.Communication overhead affects the performance of parallel computers significantly. A significant portion of the software communication overhead is attributable to message copying. We argue that it is possible to address the message copying problem at the receiving side through speculation. We show that messages display a form of locality, and we introduce the notion of message prediction for the receiving side of message-passing systems. By predicting a receive communication call before it is posted, we are able to place the required message directly into the cache speculatively before it is needed so that effectively a zero-copy communication can be achievedSpecific extensions to the ISA and the processor architecture accommodate late binding without requiring copying of the message.