On 02/03/2011 17:02, Tony Anecito wrote: > Hi Guys, > > Just to distract you from 32-bit vs 64-bit conversation Glashfish 3.1 was > released yesterday and its performance is the best I have seen for a J2EE > servlet container. They reported a 30% + improvement in almost every area and > a > less than 4 seconds startup that's even better than Tomcat 7 with my 6 core > machine.
I assume with the same apps. Parallel app deployment would improve Tomcat's start-up speed but that would require some careful changes as there are parts of the code written assuming it will be single-threaded. Not sure if the effort is worth the benefit in that case. > But I suspect for that number they were using a UltraSparc T3. They > also now provide mod_jk support so Apache can be the web server and Glashfish > the J2EE container. > > Not sure what OS they used but I suspect it was Solaris and not sure what jvm > was used. Just shows the importance in general to speeding up the J2EE > containers. To a point. With the Tomcat overhead for a Servlet around the 50 microsecond level, I'd argue there are bigger gains to be made elsewhere. Test cases that show a performance bottleneck in Tomcat are always interesting but few and far between. I did see something at JavaOne around session creation but to hit that particular bottleneck required a far from realistic test case. Even so, the bottleneck got fixed. The bottleneck that applications are most likely to hit is in database connection pooling. Borrowing and returning connections is synchronised and that can hurt apps that borrow/return a lot of connections. Work has started in Commons to improve that and it is likely to follow similar lines to the jdbc-pool currently in the Tomcat modules area that Filip has been working on. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org