https://issues.apache.org/bugzilla/show_bug.cgi?id=56798
--- Comment #7 from Bertrand Renuart <brenu...@gmail.com> --- (In reply to Chuck Caldarale from comment #6) The pool currently uses java.util.concurrent.ArrayBlockingQueue unless a "fair" implementation is requested (in which case a custom implementation is used). A stack/LIFO/MRU strategy could be built on top of a java.util.concurrent.LinkedBlockingDeque. I'm not an expert in concurrency issues but both implementations seem to make use of a single global lock for their put/get operations. We should therefore get the same performances except may be for a small penalty for the LinkedBlockingDeque due to the extra overhead of creating Nodes. I made a quick prototype by changing the idle pool into a LinkedBlockingDeque and ran about 100 concurrent threads asking for 10 thousands connections each. At first glance the difference between the two approaches is minimal. I haven't got any further though because you may already have been doing tests in that direction... -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org