A very good question. I assume you are running one of the suites of 
workloads, such as "filemicro"? If so, some workloads will finish early, 
depending on kind of application the particular workload is emulating. 
For example, the "copyfiles" workload finishes when it has copied all 
the files. That can take just a few seconds. Others will always take the 
full runtime. So think of runtime as an upper limit. If the workload 
would run "forever" then the runtime limit will stop it after <runtime> 
seconds. If the workload runs out of resources, such as files to copy, 
or files to delete, that will end the run. Finally, some workloads use 
the finishoncount flowop to stop after a set number of iterations 
through the list of flowops.

At the moment, the "running out of resources" case is not flagged as an 
error, though if you have a workload that is supposed to run 
continuously (till the runtime expires), but it stops because it ran 
out, it should be. One proposal that has been made is to have you set 
runtime to zero for workloads that are expected to run until resource 
exhaustion, and non zero for others, with resource exhaustion in the 
later case considered an error. Another similar suggestion is to have a 
mode setting in the workload to say which sort of behavior is expected. 
Comments and other suggestions would be appreciated.

Drew Wilson


Paul Monday wrote:

>I am running a very large profile with about 40 workloads...it runs a few 
>hours.  I upped the runtime to 5 minutes (300) to try to ensure the CPU is 
>balanced out.
>
>Obviously, I don't watch the terminal window during the run ;-)
>
>When I'm done, things don't add up.  For 40 x 5 minute tests, we're looking at 
>3 hours 20 minutes without the setup and teardown.  Some of my ZFS tests are 
>finishing in under 3 hours (substantially).  A few tests are core dumping, but 
>that still doesn't help bring it down.  It "seems" like I can use the creation 
>time of thisrun.f and the creation time of the .out file as guidance to 
>bracket the length of the run (hopefully that's a valid assumption).
>
>Many of the runs look to be 5 minutes...but many of them finish in under a 
>minute...and there appears to be good data in them.
>
>So, my question is...is the runtime parameter merely guidance and if the 
>filesize/threads/iosize etc... are small enough and the machine is fast 
>enough, could workloads "finish" prior to their runtime parameter?
>
>Or, is the workload always supposed to adhere to the runtime?
>
>(btw, some tests do go long in the 20  minute ballpark on slow configurations)
> 
> 
>This message posted from opensolaris.org
>_______________________________________________
>perf-discuss mailing list
>perf-discuss@opensolaris.org
>  
>

_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org

Reply via email to