On Friday, 05/05/2006 at 08:56 EST, James Melin
<[EMAIL PROTECTED]> wrote:
> Alan, Thanks for the Clarification.
>
> Based on that I shouldn't be seeing the performance I am seeing.
>
> I'm just trying to figure out what in hell can cause a performance
difference
> of 227:1 at DR vs back home. Yes. That's correct 227:1 sometimes
> slightly better, sometimes slightly worse. JITC is reporting that it is
enabled.
>
> An application that loads the initial page here in 1 second took 3:47 to
load
> at DR.
...
> If anyone has any insight into configuration issues that could cause
this, I'd
> love to hear it. I personally don't think taking Java core dumps with
> kill -3 and kill -11 of high resource Java threads at the next DR test
is going
> to reveal the source of the problem. I strongly feel that the problem
> lies in VM/VM interaction and since Linux on z/VM cannot see that layer
of
> interaction, all the Java core dumps are going to tell us is where it
was
> spending time inside Linux.

You have to have good performance analysis tools to correlate what's going
on in Linux with what's going on inside z/VM, and you need good z/VM
performance measurement tools to see what CP is doing.

Remember, too, that if everything is V=V, then you have triple paging
which is going to put quite a bit of stress on everyone.  Paging goes on
on the first level system which slows the paging of the second level
system which slows the paging of Linux even more.  This leaves less time
to actually run programs.  That do I/O.  That cause page faults.  That
cause paging.  There's a hole in my bucket.....

If the DR vendor has z/VM 4.4, and because they can run in Basic mode,
then they can put your z/VM in as V=F which will eliminate the first-level
system from the equation (for the most part) and level the playing field
somewhat.

But, again, this is a stop-gap measure, giving you breathing room to
redesign the recovery environment.  Once they move to z/VM V5 (recall z/VM
4.4 is schedule to go end-of-service in September) or go to the next level
of hardware, where you can't buy back performance with V=F, your 227:1
ratio will look like the Daytona 500 by comparison.

I know this is unlikely, but in your IP reconfiguration, did reverse-DNS
lookups break?  That will slow LOTS of Linux apps to a crawl.

Alan Altmark
z/VM Development
IBM Endicott

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to