This was apparently fixed in 6U4: http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=2147719 / http://java.sun.com/javase/6/webnotes/6u4.html.
We use Ubuntu distributions from Hardy thru Jaunty, which are tracking much more recent JDKs; 6U11 on Hardy I believe and 6U13 on Jaunty. Even my customers on RHEL 5.3 have 6U11 now, though. This bad bug is only one of the many that made anything earlier than about 6U9 unusable for us on Linux in production. I don't know if the issues are as severe on other platforms, but we could not even move from Java 5 to Java 6 in heavy load production until 6U9 -- we were loading our own javax.script jars etc. to get to Java 6 features. We had JVM crashes, 100% CPU utilization, and all kinds of other fun. I am happy to report, though, that at this point we can fire up a server with Jaunty, install the default Sun JDK packages, and run it full bore with Jetty NIO and scripting. (Jaunty's OpenJDK/Cacao support looks promising and performant, but has bizarre and plain scary bugs -- still need Sun package for now!) - R On Wed, Aug 12, 2009 at 1:46 AM, Tal Liron <[email protected]>wrote: > It can affect any user of java.nio, unless that user implements a rather > dumb workaround involving constantly destroying and recreating the > selector, which obviously creates a performance hit. As I understand it, > this workaround has been introduced into Grizzly 2.0, but I don't know > about other frameworks. Since this is a bug specific to specific > versions of the JDK on a specific platform, my guess is that some > developers would prefer not to put this workaround into their main code > branches, and delegate the problem to the user. This is especially true > when there are so many nio frameworks around, and they often compete for > performance. > > I should point out that this bug may not be easy to discover. 100% CPU > does not mean that the application stops. In fact, your application will > probably seem to run just fine. So, the only way to find this out is via > OS or VM monitoring tools. In my case, I got a friendly email from my > hosting provider asking me to stop stealing CPU cycles. :) > > Since Restlet users are heavy users of nio, I'm hoping someone else > pipes up on the mailing list with more info... > > -Tal > > On 08/12/2009 12:33 AM, David Fogel wrote: > > Hi Tal- > > > > Is it your opinion that this bug is likely to effect the > > simpleframework connector as well? If so, yikes. > > > > -Dave > > > > ------------------------------------------------------ > > > http://restlet.tigris.org/ds/viewMessage.do?dsForumId=4447&dsMessageId=2382784 > > > > ------------------------------------------------------ > > http://restlet.tigris.org/ds/viewMessage.do?dsForumId=4447&dsMessageId=2382787 > ------------------------------------------------------ http://restlet.tigris.org/ds/viewMessage.do?dsForumId=4447&dsMessageId=2382883
