Newsletter #1, February 2002

One session focused on memory management. As usual now, the emphasis was on Java!

• Pretenuring for Java Stephen Blackburn, Sharad Singhai, Matthew Hertz, Kathryn McKinley and Eliot Moss (University of Massachusetts). A scheme to detect when to allocate objects directly to old space instead of the young space. They identify allocation locations (i.e. invocations to new) that seem to create objects that live longer. They claim it works both for generational and the older-first GCs. What I found interesting in their experimental setup was that they showed GC and total elapsed times for a variety of heap sizes rather than just one. This is actually important as the heap size affects the performance of the system.
• Controlling Garbage Collection And Heap Growth To Reduce The Execution Time Of Java Applications Tim Brecht (HP Labs), Eshrat Arjomandi, Chang Li, and Hang Pham (York University). Explored how to decide on whether to collect or extend the heap in order to improve performance. They claim that if there is enough memory available, just grow the heap aggressively and don't bother collecting. However, if the memory threshold was being reached, think of performing a GC. The decision is based on how desparately out of space the system is and how much garbage was reclaimed during the last GC.
• An On-The-Fly Reference Counting Garbage Collector For Java Yossi Levanoni (Microsoft) and Erez Petrank (Technion). This paper describes an efficient, they claim, way of using reference counting (RC) for GC purposes. There is no need for synchronisation for their write-barrier as the RC updates are logged on thread-local buffers and then applied when required (this can be achieved concurrently with the program). Thanks to an interesting observation, they can get away with applying only a small number of RC updates. They use 2-bit RCs a la Wise. A backup low-priority tracing collector deals with cycles. They show that their implementation works well, however it's only been implemented in Classic JDK so it will be interesting to see how it will perform in a high performance VM. I think this is worth a read.