This is actually a bit old news, but I've been out of town.
About a week ago, we got a patch from Qingqing Zhou (and changed around in several different steps my me and Tom Lane) applied to PostgreSQL 8.1 that changes the way win32 polls for the "fake signals" used to emulate Unix behaviour and eliminates a kernel call in a very active codepath. The end result of this: Very nice speedups on Windows. In some isolated test-cases up to 40%25!!! In general cases, it's going to be much less than that of course, but it is still a significan speedup. It appears especially efficient in scenarios with lots of concurrent access, bringing win32 performance much closer to what we have on Linux.
Way to go Qingqing, and thanks also to Merlin and Andrew for quick testing work.
My own update from the Japan trip coming up in a bit...
New comments can no longer be posted on this entry.