DSM(SVM) is a memory model which was proposed at the 1980s'. DSM solved the problem thatscalability was contradict with easy programming and merge their merits. The efficiency of DSM is still aproblem. The propose...DSM(SVM) is a memory model which was proposed at the 1980s'. DSM solved the problem thatscalability was contradict with easy programming and merge their merits. The efficiency of DSM is still aproblem. The proposed solutions include prefetching and adopting new coherence model, but neither ofthem is a perfect solution. Prefetch will be less efficient when there are too many false sharing. The weak, relaxed coherence model is quite different from sequential coherence which was assumed by mostprogrammers intuitively, so programming and debugging becomes very difficult. We have proposed StrictWeak Hybrid Coherence (SWHC) model which can eliminate false sharing and provide more temporal andspatial locality, we also provide a scheme to keep sequential coherent. But SWHC model is complicate, asimplified model for Pthread share memory parallel computing model is proposed in the paper.展开更多
文摘DSM(SVM) is a memory model which was proposed at the 1980s'. DSM solved the problem thatscalability was contradict with easy programming and merge their merits. The efficiency of DSM is still aproblem. The proposed solutions include prefetching and adopting new coherence model, but neither ofthem is a perfect solution. Prefetch will be less efficient when there are too many false sharing. The weak, relaxed coherence model is quite different from sequential coherence which was assumed by mostprogrammers intuitively, so programming and debugging becomes very difficult. We have proposed StrictWeak Hybrid Coherence (SWHC) model which can eliminate false sharing and provide more temporal andspatial locality, we also provide a scheme to keep sequential coherent. But SWHC model is complicate, asimplified model for Pthread share memory parallel computing model is proposed in the paper.