Improving The Performance Of Software DSM Systems

Software distributed shared memory (DSM) systems, or shared virtual memory are advocated to be an ideal vehicle for parallel programming because of its combination of pro-grammability of shared memory and scalability of distributed memory systems. The challenge in building a software DSM system is to achieve good performance over a wide range of parallel programs without requiring programmers to restructure their shared memory parallel programs or with little modification to their sequential programs. The overhead of maintaining consistency in software and the high latency of sending messages make this target difficult to achieve.In this dissertation, to improve the performance of software DSM systems is studied by examining cache coherence protocol, memory organization, system overhead, load balancing, and communication optimization respectively.By analyzing the disadvantages of snoopy and directory-based cache coherence protocol, a lock-based cache coherence protocol for scope consistency is proposed. The unique characteristic of this protocol is applying the "home" concept to not only shared data but also coherence information. Each coherence information has a static home according to the corresponding synchronization object. To compare with directory-based cache coherence, the lock-based protocol has much less coherence overhead for ordinary read or write misses. Moreover, the lock-based protocol is free from the overhead of maintaining the directory. Based on this protocol, a simple but efficient software DSM system named JIAJIA is designed. JIAJIA employes home-based but with a novel memory organization scheme through which the overhead of address translation is eliminated and a large shared address space combined with the physical memories of all the nodes is provided. Moreover, a confidence-interval based technique is proposed to summarize the evaluation data results.Based on detailed analysis about the system overhead of software DSM systems, several optimal techniques, including read notice, hierarchical barrier implementation, cache only write detection scheme, to reduce system overheads in home-based software DSM systems are proposed. Performance evaluation results show that the performance can be improved greatly by these techniques.This dissertation proposes an affinity-based self scheduling (ABS) method for loop scheduling. ABS achieves the best performance compared with other schemes proposed in the past in meta-computing environment because of the reduction of synchronization overhead and waiting time resulted from load imbalance, and is comparable with the best scheduling schemes in dedicated environment. For iterative scientific applications, a task can be considered as a combination of computation subtask and its corresponding data subtask. Then a task migra-...