Daniel Herrington wrote:
> On Tue, Oct 18, 2011 at 1:35 PM, Paul Heinlein <heinl...@madboa.com> wrote:
>
>
>> On Tue, 18 Oct 2011, Daniel Herrington wrote:
>>
>>
>>> All,
>>>
>>> We're working an issue with extreme latency on one of our
>>> application servers. The lead tech, who hasn't established much
>>> credibility, keeps saying he wants to bounce the Sun Solaris
>>> servers, as they have been up for 169 days. He feels that may be the
>>> cause of the issue. I highly doubt it as the System Administrators
>>> are saying that resources are available for the application. What's
>>> the recommended reboot cycle for Sun Solaris servers?
>>>
>> I haven't administered any mission-critical Solaris boxes for several
>> years, but I never had to reboot Solaris to solve an application-level
>> problem. I suspect truss or dtrace can identify the source of the
>> latency, if it actually is caused by Solaris.
>>
>>
>>
> I don't think the latency is coming from the OS, but who knows. At this
> point the Scheduler (CA Inc) is seeing it in the OCI calls, but that just
> means it's downstream from them. Oracle is saying they can't find anything
> in the logs, and so I've got a whole lot of shrugging shoulders. The Storage
> guys say the OS can tell is they are having resource issues, and the OS guys
> are saying they don't see any resources spikes. The only thing I do know is
> that no one has a clue. To solve the problem we 'roll around the horn' so to
> speak on the RAC environment. Latency disappears after that. it smells like
> an Oracle issue, but at this point I'm stuck.
>
Not sure this is going to help much, but just in case
IRIX 6.4 (SGI MIPS)
Oracle 8.0.4
What appear as a latency issue was that Oracle didn't like the way SVR4
systems counted time, and processes (it was a know bug but Oracle's fix
was to move to 9.x which wasn't supported on those offending OS's). The
known bug, you see, was found on other SVR4 systems as well, and
Oracle's known work around was to periodically power cycle the
machine(s), resetting the counters in question. In our case it was <28
days (we made it 21 so it could always happen on Sundays), on some SVR4
boxes it was <280 days ... that was dependent on how the OS counted time.
Hope that helps
Regards
Fred James
_______________________________________________
PLUG mailing list
PLUG@lists.pdxlinux.org
http://lists.pdxlinux.org/mailman/listinfo/plug
