On 13/12/2013 13:12, Bob DeRemer wrote: > We have been profiling this in YourKit and as we increase the number of > websockets, naturally we see the CPU utilization increase. The concern > is that we’re seeing this at what seems to be a fairly low number of > websockets. The YourKit profiling showed 41% of the time being spent in > the following code. My apologies for the formatting. I can provide an > attachment out-of-band if it would be helpful.
I've seen similar figures from YourKit when testing with just a single connection. There does seem to be a lot of time spent in the NIO Poller code. When I looked at it last time nothing immediately obvious jumped out at me in terms of something that could be changed to improve things. It needs some more time spent on a deeper analysis. That is on my TODO list. If any NIO experts want to take a look and make some suggestions that would be great. As an aside, as much as I like YourKit, you do have to be careful (as you do with any profiler) when stress testing that the profiler isn't distorting the results. I have been led down the wrong path a few times when I wasn't careful enough. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
