cluster's blog

As I posted in my previous blog entry work had begun to improve JNodes memory management. We had sucessfully implemented our initial design of using a single heap instead of multiple linked heaps, and utilizing a special structure for effecient storage of small objects. The problem I found after the new system was working was the ability to tell wether or not anything was actually improved, especially when I started to reduce the memory managers locks to a finer grained level. In theory, the allocation speed would be greatly increased, but we had no simple way of telling as multiple attempts at getting benchmarking applications to run on JNode had failed. Without this it was impossible to tell what sort of affect our new system was having on the runtime performance of JNode, which is as much or more important than the allocation speed of the memory manager.

Implementation for new memory manager has begun for JNode. The new design is being planned to lay a foundation for a more concurrent memory manager, and a more effecient garbage collection. In order to do this, the new memory manager and its heap designs are going to be developed in a way to allow the most concurrency through finer grained heap structures. I'll talk first though about the changes being made to go from JNode's current MM to the new MM.