At 01:05 PM 11/2/2005, Quanah Gibson-Mount wrote: >With this latest change, I found HEAD to only be 1.5% slower than OL2.3.11 for >READs. I found no performance gain at all with HEAD as I increased the number >of querying clients -- Rates dropped off the same as they did in 2.3.11. So >this may prevent connections from being deferred, but it doesn't appear to >allow any more work to be done, at least where READs are concerned.
I note that the intent of the patch is not "to allow any more work to be done" but to prevent blocking in the dispatcher which can cause no new work to be done. Or to put it another way, its to make better the worse case. Kurt
