II-New: A Cluster of Nodes with 32 Cores and 256-GB Memory to Enable Many-Core Systems Research and Education

点击翻译按钮获取中文摘要

Research and education in many-core computing systems are of importance to the NSF CRI program as well as to the research community. This project targets performance, energy consumption, and scalability of many-core systems, which are important for the computer industry. The team is committed to releasing the research artifacts of the project as open-source software to be used by the research community as well. This project will benefit graduate student research and help educational activities in undergraduate and graduate curricula. The project will support outreach activities sponsored by various centers at Purdue University via the involvement of the team in Purdue Computing Research Institute's High Performance Computing workshops, for example.This infrastructure will support research and education efforts in multiple areas: computer architecture, compilers, high-performance cloud computing, and run-times for managed languages. Computer architects will explore optimizations for performance, programmability and power of many-core architectures, on-chip networks, and disk optimizations. Compiler writers will explore shared memory optimizations and their scalability targeting shared-memory applications for distributed memory machines, and techniques to transform seemingly irregular memory access patterns into regular and parallel computations and memory accesses. Run-time researchers will pursue parallel garbage collection of large garbage-collected heaps and associated scalability issues. High performance computing researchers will explore the performance overhead of virtualization and cloud computing for cluster workloads, along with mechanisms for reducing overhead.