This paper describes a new load balancing strategy for massively parallel computers called Probabilistic strategy with Neighbourhood Synchronization (PNS). The proposed strategy differs from fully distributed aproaches which require a high interprocesor communication overhead when the number of processors becomes large. This load balancing strategy is based on a dynamic process mapping algorithm which uses only status information from neighbour nodes and takes into account the information lags in message-passing systems for estimating the system load. After the description of the algorithm, some experimental results obtained from a real implementation of the load balancing strategy on a transputer-based multicomputer are presented and discussed.
PNS - A DYNAMIC MAPPING STRATEGY FOR MASSIVELY-PARALLEL COMPUTERS
Sergeyev Y;TALIA D
1994-01-01
Abstract
This paper describes a new load balancing strategy for massively parallel computers called Probabilistic strategy with Neighbourhood Synchronization (PNS). The proposed strategy differs from fully distributed aproaches which require a high interprocesor communication overhead when the number of processors becomes large. This load balancing strategy is based on a dynamic process mapping algorithm which uses only status information from neighbour nodes and takes into account the information lags in message-passing systems for estimating the system load. After the description of the algorithm, some experimental results obtained from a real implementation of the load balancing strategy on a transputer-based multicomputer are presented and discussed.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
