Cooperative Memory Management for Multi-core Systems

While a conventional program uses exactly as much memory as it needs, the memory use of a garbage-collected program can be adjusted by changing the size of the heap used by the garbage collector. This difference provides challenges and opportunities for both the application and operating system. Garbage-collected application can improve their performance by adapting their memory demands in response to the changing amount of available memory in today's multicore, multiprocessor machines. Because applications' demands are not fixed, operating systems must also adopt how to best allocate memory between the concurrently executing processes.

We first present our adaptive system that allow garbage-collected applications, working both independently and in conjunction with other executing applications, to adjust and optimize its memory usage. As we show using a series of diverse workloads, this system improves overall throughput by up to a factor of 4.5 without ever degrading performance. We also descibe how operating systems can be modified to better account for this new application behavior. By improving how the OS manages memory for concurrently executing applications, we can make application performance more predictable and decrease throughput time. Taken together, this work shows why memory management must change in multiprogramming, multi-core environments and how these changes can be made.

Greg Steffan

Last modified: Wed Aug 26 18:21:23 EDT 2009