On 11/03/2015 08:07, Rémy Maucherat wrote: > 2015-03-10 23:35 GMT+01:00 <ma...@apache.org>: > >> Author: markt >> Date: Tue Mar 10 22:35:19 2015 >> New Revision: 1665736 >> >> URL: http://svn.apache.org/r1665736 >> Log: >> Stop re-using the SocketWrapper >> With the introduction of upgrade and non-blocking, I/O can occur on >> non-container threads. This makes it near impossible to track whether a >> SocketWrapper (== connection) is still referenced or not, making re-use a >> risky proposition. >> >> Modified: >> tomcat/trunk/java/org/apache/tomcat/util/net/AprEndpoint.java >> tomcat/trunk/java/org/apache/tomcat/util/net/Nio2Endpoint.java >> tomcat/trunk/java/org/apache/tomcat/util/net/NioEndpoint.java >> tomcat/trunk/java/org/apache/tomcat/util/net/SocketWrapperBase.java >> > > I had tested enabling reuse of the wrapper with NIO2, and it didn't change > the ab results. Same for using the useCaches flag, so there's probably no > point either reusing the other objects in endpoint.
I while back I did some GC testing under load and I think some of those caches were introduced to reduce the GC churn. Given none of them are obviously causing problems my instinct is to leave them as is for now. My current TODO list is: - fix this APR crash - fix the remaining bugs that impact 8.0.x - do an 8.0.x release - do some performance testing I can add looking at the performance and GC benefits (or not) of these caches to that last item although I'd be more than happy to if someone beat me to it. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
