Ram Rajamony - Research Summary

Future research plans

• Prescriptive performance tuning

  This is a new approach that I have developed for designing tools to improve application performance. In several domains, a tool based on my approach can prescribe changes to improve performance, as opposed to just describing the program behavior.

• Optimally synchronizing loops

  When a compiler parallelizes sequential loops for execution on a shared memory multiprocessor, it may need to add synchronization to enforce data dependences. I have developed a set of algorithms to add the minimal synchronization that enforces all data dependences.

• Compiler support for irregular applications on distributed shared memory

  When executing irregular applications on a distributed memory machine, a message-passing program performs better than a shared-memory program running on top of a software distributed shared memory (DSM) system. This research shows that a modest amount of compiler support for aggregating messages enables a software DSM system to close the performance gap, while retaining the advantages of a shared-memory programming model.

• Software distributed shared memory systems

  My research contributions in this area include the first detailed comparison of lazy release consistency and entry consistency, two relaxed memory models used in software distributed shared memory systems. I established that the programmer annotations required by entry consistency for correctness do not translate into improved performance.

Back home