Research of Parallel Computing Model based on Multi-core Processor
Junfeng Wang · Intelligent Computer and Applications · 2011
As there is not a universal model in parallel computer architecture,the paper analyzes some existing typical models,comparing them in synchronization,communication,parameters,then proposes a new model: mzLogGP based on LogGP.After that,the paper analyzes and tests the parallel process using the MPI parallel algorithm in the environment of non-exclusive nodes on the computing resources and existing network congestion,and calculates the computational overhead and communication overhead by increasing two parameters: network congestion and layers of memory hierarchy.In the end,by comparing the measured time with the predicted time,the paper shows that with the increase in the number of nodes,the system error decreases.The results demonstrate that the new model can improve running performance of parallel application in the platform of multi-core processor cluster,and has better scalability.