Hi Martha: If it is at all possible for you to do I suggest that you
work with your Oracle DBA to identify viral SQL statements.  If I hadn't
experienced this myself with a multiple virtual CPU Linux Oracle
environment  I wouldn't have believed it - but one miscreant SQL
statement took 40% of a z9BC IFL. Tuning the application while it was
growing in demand, size of DB, and code paid off big time reducing from
6 to 4 IFLs. So it wasn't a one time deal but saved big $$.

There is a gadget in Oracle which tells you the longest performing SQL
statements.
Unfortunately the fixes involve someone changing application code.
David Kreuter

-------- Original Message --------
Subject: Virtual cpu question
From: Martha McConaghy <[email protected]>
Date: Wed, September 29, 2010 3:45 pm
To: [email protected]

We have an interesting problem. We have a big Oracle database (part of
our
ERP) running on SLES 10. At the end of October, we will be doing our
first
full student registration with the new system, so we have been running
some
load tests using a program named "grinder". I'm not sure how realistic
the
tests are, but its what we have to work with.

The system is running in an LPAR on a z9 with 3 IFLs. I have 3 virtual
IFLs
defined for the machine, along with 6G of memory. When they run their
tests,
the memory and I/O seem fine. The problem is that from the Linux point
of
view, the 3 virtual processors get pegged at 100% load. I can see from
the
CP stats that VM is running very high (cpu load), but is not at 100%.
So,
it seems to be the virtual processors that are throatling it.

As an experiment, I moved the virtual machine over to our older z990, to
an LPAR with 5 real processors. These are slower than the z9, but there
is
much less load overall. I defined 5 virtual processors to the virtual
machine
and we ran the tests again. The difference was dramatic. The virtual
processors never reached their max of 100%. The throughput of the
transactions was much improved. So, despite the slowness of the real
processors, adding 2 more virtual processors made a big difference.

My understanding has always been that you should not define more virtual
processors than real processors for a virtual machine. Does this still
hold
when dealing with Linux guests? Does anyone have any suggestions on how
to
deal with the problem? I'd like to keep the server over on the z9, but I
don't have any more real processors to add to the LPAR.

Martha

----------------------------------------------------------------------
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
----------------------------------------------------------------------
For more information on Linux on System z, visit
http://wiki.linuxvm.org/

----------------------------------------------------------------------
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
----------------------------------------------------------------------
For more information on Linux on System z, visit
http://wiki.linuxvm.org/

Reply via email to