Extending the Limits of Large-Scale Shared Memory Multiprocessors

点击翻译按钮获取中文摘要

The objective of this research is to substantially improve the productivity of programmers writingapplications for petaflop-scale systems by using programmer defined light-weight transactions as thesingle abstraction for expressing parallelism, delineating communication, reasoning about memoryconsistency, providing failure recovery, and allowing performance optimization. Transactions as thecentral abstraction for designing and programming parallel systems leads to a shared memoryprogramming and memory coherence model called Transaction Coherence and Consistency (TCC).Transactions simplify parallel programming by providing a way of writing correct shared-memoryprograms without threads, locks and semaphores. TCC systems provide high performance communicationand synchronization with support for hardware mechanisms that can keep memory coherent andconsistent based on programmer-defined transactions.To achieve the research objective, this research program will focus on five activities. First, the researcherswill develop new abstractions that use transactions to provide a shared memory programming model thatmakes it much easier to analyze and optimize application performance. Second, the researchers willdevelop performance monitoring systems that make use of transactions to detect performance bottlenecksand to provide intuitive feedback to programmers. Third, the researchers will use the transaction basedprogramming model to implement compiler-based static and dynamic feedback-directed optimizationsthat automatically detect and eliminate performance bottlenecks and extend the scalability of transactioncoherency to 105 processors. Fourth, the researchers will use transactions to optimize the performance ofparallel storage I/O. Finally, the researchers will develop simulation and emulation technology that willenable us to experiment with petaflop-scale systems that support light-weight transactions before they areavailable.Broader ImpactsThe broad impact of this research is to use transaction-based parallel programming to educate and enablea new class of parallel software developers who can implement parallel software with the same facilitythat sequential software is written today. Enabling parallel software development will be critical toadvancing computing performance from desktop applications to large-scale scientific and commercialapplications. While parallel processing has been essential for large-scale machines for a while, recentannouncements by Intel, AMD and IBM demonstrate that it will soon be critical for desktop applicationsas well. To educate students, other researchers, and industry about the benefits of transaction-basedparallel programming, we will incorporate transactional programming concepts in the parallelprogramming curriculum and make transaction-based applications available to the wider scientificcommunity. The researchers expect that releasing a suite of optimized transaction-based applicationsalong with simulation technology will be instrumental in encouraging other researchers to experimentwith and explore the benefits of transactions. To further promote the use of transaction-based parallelprogramming we will organize a tutorial or workshop at a major scientific computing conference that willcover the principles and experience of programming with transactions.