John,
I was collecting sar at one-minute interval and that is how I was able
to determine that the CPU remained pegged for 6-8 minutes. In production
the middle-tier is load-balanced via Cisco's CSS switch. What I do not
understand is that if the issue is with CPU resources then why the error
appeared when there were ample CPU available on the server.
From: John Kanagaraj [mailto:john.kanagaraj_at_hds.com]
Sent: Friday, November 11, 2005 7:01 PM
To: Hameed, Amir; oracle-l_at_freelists.org
Subject: RE: Mercury LoadRunner error message
Amir,
Keep in mind that 'sar' averages out the figures over the
collection period (usually 20 mins?) - peaks and valleys thus get lost.
The error cannot be clearer - the mid-tier does not seem to be able to
respond within a given time. I would suggest upgrading the middle-tier
(or at least load balancing it with another box). In a live situation,
you don't want a single mid-tier box to become the weakest link, right?
John Kanagaraj <><
DB Soft Inc
Phone: 408-970-7002 (W)
Co-Author: Oracle Database 10g Insider Solutions
http://www.samspublishing.com/title/0672327910
- The opinions and facts contained in this message are entirely
mine and do not reflect those of my employer or customers **
From: oracle-l-bounce_at_freelists.org
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Hameed, Amir
Sent: Friday, November 11, 2005 3:29 PM
To: oracle-l_at_freelists.org
Subject: Mercury LoadRunner error message
I need some help on troubleshooting an issue with the Mercury
LoadRunner. We are running a LoadTest with Mercury LoadRunner 7.8
against an RAC-enabled Oracle 11i (11.5.9) suite. This is a three-tier
architecture with three RAC nodes on the backend running Oracle
instances, administration and concurrent processing servers and one v880
server on the middle-tier running forms and Apache servers. Some of the
LoadRunner connections have been dropping with "Error in reading from
Socket or Server not responding". The middle tier has 6 CPU and 12GB RAM
and we are driving approximately 120 forms sessions. When the first time
this message appeared, I looked at "sar -u" and "sar -q" statistics and
discovered that the CPU idle was almost zero and the run-queue length
was b/w 12-14 with queue 100% full. This condition had sustained for
about 6-8 minutes and thereafter the CPU got released. The wait time
defined within the LoadRunnder is 2 minutes and I have been told by our
testing folks that if a LoadRunner virtual user connection is not
established during this period then the above message is encountered.
However, there a few drops in the later part of the afternoon where
neither the CPU was pegged nor the run-queue was full. If anyone has
seen this issue then could you please shared with me the reason under
what circumstances the LoadRunner terminates a connection with "Error in
reading from Socket or Server not responding" message.
Any help will be appreciated.
Thanks
Amir
--
http://www.freelists.org/webpage/oracle-l
Received on Fri Nov 11 2005 - 18:34:40 CST
