Hmmm, the apache mailing list won't let my current email post a reply.
Trying again.
-Akara
-------- Original Message --------
Subject: Re: Parsing Olio runtimeStats
Date: Wed, 05 May 2010 14:17:35 -0700
From: Akara Sucharitakul <[email protected]>
Reply-To: [email protected]
Organization: Oracle
To: [email protected]
CC: akara.sucharitakul <[email protected]>, Shanti Subramanyam
<[email protected]>
References:
<[email protected]>
Let me try to address this below:
On 05/05/10 12:12, Vasileios Kontorinis wrote:
Akara and Shanti hi,
I am parsing the runtimeStats from the driver.log file and I run
into this interesting issue.
In the output for _steady state_ I get "-" for the response time when
there are no successful operations since the last time runtimeStats
were printed.
This can happen for two reasons:
1) The interval for printing the runtimeStats is small (in my case
5secs) and some operations, especially the ones that take long and
have small frequency in matrix (add event, add user) just never happened.
Yes, if the number of successful operations in that period is 0, most
calculations will be a divide by 0. That's why it is printing a '-'.
2) There are a bunch of requests going on and none of them is
successful. This can happen when there is no sufficient memory in the
machine. In that case there is lots of swapping, the cpu goes to 100%
utilization and all the operations time-out.
Is there any small change I can do to distinguish between the two
cases? Maybe printing a small "t" when there are many requests timing
out, so that I can distinguish between the two cases? Any ideas are
welcome.
There is the error count in the runtime stats that tell you about error
cases. This would also include timeouts. But the problem is not that
simple. The real problem is not the operations timing out but rather the
operations waiting that would time out. And we cannot distinguish
between these waiting operations and other operations waiting just with
slow response time.
One more thing. I use the variableLoad setting and in order to know
how many users are simulated while the benchmark run, I parse log.xml
looking for "Active threads: ". The problem is that log.xml is big
especially when many requests timeout. Is it easy to also log the
changes of active threads in the drive.log file ??
Please look into log configuration in $FABAN/config/logging.properties.
You should be able to make certain loggers log to a particular file. I
don't have the detail off the top of my head.
-Akara
