The advent of multi-cores drives the biggest revolution of software development - parallel programming. Writing efficient parallel programs remains challenging. Resource contention at the cache line level, normally invisible to programmers, can cause false sharing. False sharing can dramatically degrade performance, as much as an order of magnitude, and severely affect scalability of programs. The hardware trend of building more cores or using larger cache line size increases the prevalence of false sharing problems.

We develop the first tool to correctly and precisely pinpoint the exact cause of false sharing. This tool is also able to generalize from the current execution to accurately predict latent false sharing that may appear in a slightly different execution environment. Rewriting a program to fix false sharing can be infeasible when source code is unavailable, or undesirable when padding objects can increase excessive memory consumption or further worsen runtime performance. To resolve this problem, we further provide the first runtime system that automatically eliminate false sharing inside multithreaded parallel applications without programmer intervention.

Tongping Liu is a PhD student in the school of Computer Science at the University of Massachusetts Amherst. His research spans runtime systems, operating systems, programming languages, and distributed systems. His primary research goal is to improve the reliability and performance of parallel software.