Adapting wave-front algorithms to efficiently utilize systems with deep communication hierarchies
- Los Alamos National Laboratory
Large-scale systems increasingly exhibit a differential between intra-chip and inter-chip communication performance. Processor-cores on the same socket are able to communicate at lower latencies, and with higher bandwidths, than cores on different sockets either within the same node or between nodes. A key challenge is to efficiently use this communication hierarchy and hence optimize performance. We consider here the class of applications that contain wave-front processing. In these applications data can only be processed after their upstream neighbors have been processed. Similar dependencies result between processors in which communication is required to pass boundary data downstream and whose cost is typically impacted by the slowest communication channel in use. In this work we develop a novel hierarchical wave-front approach that reduces the use of slower communications in the hierarchy but at the cost of additional computation and higher use of on-chip communications. This tradeoff is explored using a performance model and an implementation on the Petascale Roadrunner system demonstrates a 27% performance improvement at full system-scale on a kernel application. The approach is generally applicable to large-scale multi-core and accelerated systems where a differential in system communication performance exists.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC52-06NA25396
- OSTI ID:
- 971329
- Report Number(s):
- LA-UR-09-06298; LA-UR-09-6298; TRN: US201004%%91
- Resource Relation:
- Journal Volume: 37; Journal Issue: 9; Conference: IEEE International Parallel & Distributed Processing Symposium ; April 19, 2010 ; Atlanta, GA
- Country of Publication:
- United States
- Language:
- English
Similar Records
Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)
Multi-petascale highly efficient parallel supercomputer