CAREER: Deterministic Shared Memory Multiprocessing: Vision, Architecture, and Impact on Programmability

点击翻译按钮获取中文摘要

Software innovation typically relies on performance improvements ofthe underlying hardware. However, technology limitations hinderfurther significant progress in single-thread performance. Therefore,the software industry has the immense problem of rethinking itssoftware development process and techniques to adopt multicoresystems. Popularizing parallel programming is a Grand ResearchChallenge for the systems community [CRA]. Being able to leverage thefull potential of multicores would put us back into exponential growthof usable performance as well as lead to significant power savings.One of the main reasons why parallel programming is hard is thatparallel code in current multicore systems can executenondeterministically. Each time a multicore executes a parallelapplication, it can produce a different output even if supplied withthe same input. This frustrates debugging efforts and limits theability to properly test parallel code, becoming a major obstacle towidespread adoption of parallel programming. This project poses broad intellectual questions with far-reachingimplications in modern computer systems: Can nondeterminism be removedfrom shared-memory multiprocessor systems without degradingperformance? What are the trade-offs in designing deterministicmultiprocessor systems? What are the implications and uses ofdeterministic behavior in programmability? The PI plans to answer thesequestions by devising efficient, general purpose, fully deterministicshared memory multiprocessor systems and demonstrating that they canenable significant changes in how parallel programs are written,tested and deployed. An integral part of the concurrency challenge iseducation, so this project also aims to develop a graduate andundergraduate curriculum that will teach students about concurrencyprinciples and practical parallel programming.