On 10/13/11 5:31 PM, Craig Setera wrote:
Assuming that the errors and warnings are not a big deal, can you suggest
any reasons that we would have connections stacking up and timing out? The
use case is likely heavier on write than would be standard for LDAP, but it
seems the failure threshold (number of connections) is very low. This is
hosted on a multi-core machine (virtual machine) and when it gets bad, our
operations people say that a single core of the machine is pegged at 100%
CPU while others are essentially idle. Are there parts of Apache DS that
have thread affinity and would be "stuck" to a single processor?
Ok, 1.5.5 is pretty old, and depends on MINA-2.0.0-M6. This version of
the NIO framework sometime gets stuck with 100% CPU (it's a JVM bug)
when some session are closed just after having been opened (basically,
the selector loop get crazy and eats 100% CPU looping).
It might be the reason why you see this 100% CPU for some sessions (as
we have one selector per CPU, the other sessions might be ok).
I would suggest you switch to at least 1.5.7, but even better to 2.0.0-M3.
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
