2015-01-15 17:03 GMT+01:00 Mark Thomas <ma...@apache.org>: > On 15/01/2015 14:28, r...@apache.org wrote: > > Author: remm > > Date: Thu Jan 15 14:28:13 2015 > > New Revision: 1652108 > > > > URL: http://svn.apache.org/r1652108 > > Log: > > Fix use of the semaphore (it seems equivalent to not using it). This > fixes the ab corruption I was experiencing. > > Thanks. That fixes the performance issue I was seeing (which was caused > by an exception during socket close so the lower maxKeepAliveRequests > the more obvious the problem became). It certainly drives home the > message that you never, ever want exceptions routinely occurring on the > critical path. > > For the ab results, I am now at (8 to trunk before latest refactoring to trunk): NIO1: 50k -> 60k -> 57k APR: 67k -> 67k -> 62k NIO2: 65k -> 65k -> 60k
So I see a small difference with this latest refactoring. Rémy