Bill Shannon wrote:
> Marty Itzkowitz wrote:
>> Interesting problem. I've used disk rattle as a measurement of io
>> activity before
>> there were such tools for measurement. It's crude, but effective.
>>
>> To answer your question: you could try er_kernel. It uses DTrace to
>> do statistical callstack sampling, and is described on our kernel
>> profiling page <http://dpweb.sfbay/products/analyzer/kernel.html>.
>> That page is woefully out of date, but is basically correct w.r.t usage.
>> We test it on Nevada and S10, and S9.
>>
>> Roch Bourbonnais (PAE), who did most of the development, used it to
>> track
>> down the root cause of a similar-sounding problem in VxFS. See the
>> slides
>> towards the end of the presentation on Kernel Profiling
>> <http://dpweb.sfbay/products/analyzer/kprof.pdf> dated June, 2002.
>>
>> The best version to use on a late snv is the one from our
>> nightly-build <http://dpweb.sfbay/products/analyzer/access.html#nightly>
>>
>> If you have any problems, please contact Roch and/or me;
>> if you give it a try, please report your experience back to our interest
>> alias, [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]> .
>
> I looked through the page and the presentation and it wasn't clear to me
> that this was going to give me information on which file was being
> accessed or which process was doing the accessing. I really want
> "tracing",
> not "profiling". Statistically, the disk isn't being accessed at all.
It should give you information on what callstacks in the kernel are
triggering
the disk usage. You should see something with the same periodicity as the
rattle you hear.
Marty
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
