NGS: A Computing Environment Based on Hardware Transactional Memory

点击翻译按钮获取中文摘要

Abstract:Charles Leiserson, CNS-0305606Parallelism introduces more opportunities for unwanted nondeterminism in programs, making them hard to debug and impossible to write reliable regression tests for. Consequently, parallel computing has emerged as a niche technology suited only for software developers who are willing to cope with the complexities of concurrency, because performance is everything for them.In 1993, Herlihy and Moss{58} proposed transactional memory as an alternative mechanism for enforcing atomicity, since it allows the user to avoid many of the complications of locking. With transactional memory, a program can read and modify multiple, disparate memory locations as a single atomic operation, much as occurs within a database transaction. The trend has been towards software transactional memory(STM) {98,57,56}, the overhead of which discourages the use of transactions. The project contends that the "transactions everywhere" approach can simplify parallel programming dramatically and that hardware support can make overheads negligible. To test the project hypothesis, they plan to develop a hardware specification of "scalable" HTM with a software simulator, a linguistic framework for C-like languages, complier support "implicit" transactions, debuggers and other software-engineering tools, theoretical foundations for semantics and performance, and multithreaded applications that demonstrate the advantages of an HTM computing environment.