Abstract

With the advent of multicores, there is demand for monitoring parallel programs for a wide range of applications including those related to improvement of performance and reliability. Debugging tools such as data race detection, tools such as DIFT (dynamic information flow tracking) for detecting software attacks, techniques such as speculative parallelisation that strive to expose parallelism and improve performance, are examples of applications that require a parallel program to be monitored during runtime. While each of the above monitoring applications are quite different in their purpose and implementation, they all share a common requirement in the context of monitoring a parallel program—the need to detect and react to interprocessor shared memory dependences (ISMDs).

In this talk, we propose a simple architectural extension in the form of ECMon: support for exposing cache events to the software, in effect, efficiently exposing the ISM Ds to software. We demonstrate how a wide variety of monitoring applications, including those for security attack detection, those that allow speculation past barrier synchronisations, and those for ensuring sequential consistency in a machine with a relaxed memory model, can be implemented efficiently using this support.

Bio: Vijay Nagarajan recently joined the University of Edinburgh as a lecturer. He obtained his PhD in Computer Science from the University of California, Riverside and Masters in Computer Science from University of Arizona. His research interests span the areas of computer architecture, compilers, and software engineering and has published in some of the premier conferences and journals in these areas. More information about his research can be found at http://homepages.inf.ed.ac.uk/vnagaraj/