CSR: Small: Software Transactional Memory for Real-Time Systems

University Of North Carolina At Chapel Hill, Chapel Hill NC

Investigators

Abstract

When someone driving a car presses the brake pedal, they expect the anti-lock brakes to react in time to avoid a problem. A car actually contains many computers that run programs that control braking, steering, engine power, etc. These programs must perform their control operations by specific deadlines, often measured in tiny fractions of a second, to ensure safe operation. Systems like those in a car belong to a category of hardware and software designs called "safety-critical real-time systems" because they must complete their work by real deadline times that matter for safety. Systems in this category require some form of certification that provides the necessary level of assurance that catastrophic consequences, such as loss of life, are avoided. Most car buyers would likely never give certification a passing thought, but it is a critical part of the automotive design process. This project is directed at developing infrastructure that can aid in the creation and certification of safety-critical real-time systems. This infrastructure will be provided in the form of a concept called real-time software transactional memory (STM). The purpose of real-time STM is to allow tricky program-certification issues, which arise in complex software systems in which programs share data, to be handled automatically with little programmer effort. The proposed real-time STM framework will ease overall certification efforts and costs in the development of complex safety-critical real-time systems, which are increasingly becoming part of our society and daily lives. In developing this framework, multicore processors will be a primary emphasis. The advent of multicore technology has created serious certification issues in the avionics and automotive industries. The real-time STM to be developed will be specifically disseminated to research collaborators in these industries to evaluate in their systems. The software and data produced will also be made available for public download by other interested parties. Instructions for doing so will be provided at http://jamesanderson.web.unc.edu/research/. The results obtained in this project will be disseminated to the public through lectures at universities and international summer schools, presentations at conferences and workshops, and public demonstration events. Computer science as a field is suffering from a serious shortage of women at all levels and in particular at the graduate level and above. Several female graduate students will be directly involved in this project