Postdoc: A Hierarchical Methodology for SMP Clusters

0203584Moret，新墨西哥贝尔纳大学实验计算机科学的博士后助理：SMP集群的分层方法到目前为止，并行计算中隐含的高性能计算的承诺已经在有限的应用范围内实现，但大型分布式存储系统中的并行受到消息传递成本的阻碍。对称多处理器(SMP)集群将分布式内存与真正的共享内存访问结合在一起，为有效解决复杂、不规则的优化问题提供了潜力，并降低了消息传递成本。建议的研究是利用计算生物学扩展高性能SMP计算的工作，为SMP集群开发一种分层方法。这位博士后研究助理将使用高性能计算生物学研究中的问题和算法作为测试案例，开发和测试适用于未来多级并行体系结构的方法。

0203584Moret, BernardUniversity of New MexicoCISE Postdoctoral Associates in Experimental Computer Science: A Hierarchical Methodology for SMP ClustersThe promise of high-performance computing implicit in parallel computation has so far been realized in a limited range of applications, but parallelism in large distributed-memory systems is hampered by the cost of message passing. Symmetric multiprocessor (SMP) clusters combine distributed memory with true shared memory access and offer the potential for effectively solving complex, irregular optimization problems with reduced message passing costs. The proposed research is to extend work on high-performance SMP computation, using computational biology, to develop a hierarchical methodology for clusters of SMPs. The postdoctoral research associate will develop and test the methodology, suited to future multilevel parallel architectures, using problems and algorithms from research in high-performance computational biology as test cases.