SHF: Medium: Programmable Monitoring Framework for Multicore Systems

University Of California-Riverside, Riverside CA

Investigators

Abstract

The advent of multicore processors has introduced new opportunities for achieving increased software performance, reliability, security, and availability. However, powerful dynamic execution monitoring capabilities are required to realize these opportunities. This project addresses the challenges of developing a Dynamic Binary Translation based monitoring framework for parallel applications running on multicore systems. The programmability of the framework will enable realization of benefits in achieving enhanced performance, reliability, security, and availability. Some of the instrumentation code required in context of parallel applications must be executed by a core in response to events that involve other cores. In particular, events relevant to many performance, reliability, and security related tasks correspond to the manifestation of interprocessor data dependences due to updates of shared memory locations by multiple cores. Based upon this observation programmable architectural mechanisms will be provided that not only enable the detection of interprocessor dependence events but also enable the triggering of the execution of application specific monitoring code. This project will then employ these mechanisms for improving performance via speculative parallelism, enabling debugging via a novel strategy of execution suppression, improving reliability via an approach that allows applications to automatically recover from failures, providing security via dynamic detection of mutating viruses, and software availability via dynamic updates.