So if you're not seeing the CPU spikes anymore, it's hard to guess
what's going on -- but given that the glassfish threads are all idle in
the latest stack dumps, it seems likely that the problem is elsewhere.
The most likely culprit seems to me to be the network -- I did notice in
some of the stack dumps that a few glassfish threads were blocking
writing data back to the clients, which would normally only happen if
there is a network issue. I wonder if running tcpdump/snoop on both the
client and server machines would show any delays. Or seeing how much ftp
thoughput you get during those 30-second intervals. Or even simple ping
tests (or traceroutes) could show a long time delay.
We have seen on some platforms (like Windows 2003) that the OS fails to
send us the appropriate signal when a request is pending; I suppose
something like that could be going on. That might also explain why it
started happening suddenly, since an automatic OS update might have
introduced the problem. But that seems much less likely to me; it would
be a big OS bug (Windows 2003, not being a server-targeted OS, didn't
necessarily cause so many difficulties with its issue).
-Scott
On Wed, 2009-03-25 at 12:28 -0700, glassfish_at_javadesktop.org wrote:
> Hi Scott,
>
> Just to clarify, the original issue for creating this thread was investigating increases in CPU and interrupts / context switching that occurred during non-peak traffic. Now, however, our graphs are unremarkable. We're barely using any CPU ( <8% ) and the interruptes / context swtiching graphs don't show any oddities, yet server response times can be 30s+. It's confusing because the codebase is the same we had run for a week with much higher traffic without problem.
> [Message sent by forum member 'rwillie6' (rwillie6)]
>
> http://forums.java.net/jive/thread.jspa?messageID=339036
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe_at_glassfish.dev.java.net
> For additional commands, e-mail: users-help_at_glassfish.dev.java.net