dtrace discuss - Aug 2009 - disk service time using dtrace

Hi,

           I am trying to figure out the service time distribution for disk
reads using dtrace. I am using Solaris Community edition, with UFS as the
filesystem. I have written simple scripts which use fbt, but am not getting any
consistency in my results. The biggest hindrance seems to be the caching
effects. Any suggestions on how to get an average even though caching effects
are there?

Thank you,
Vish
-- 
This message posted from opensolaris.org

G''Day Vish,

On Tue, Aug 11, 2009 at 03:39:01PM -0700, vhiz wrote:
> Hi,
> 
>            I am trying to figure out the service time distribution for disk
reads using dtrace. I am using Solaris Community edition, with UFS as the
filesystem. I have written simple scripts which use fbt, but am not getting any
consistency in my results. The biggest hindrance seems to be the caching
effects. Any suggestions on how to get an average even though caching effects
are there?
Which cache do you mean?  If you use the io provider, you should be close enough
to disk access that you''ll only be affected by the on-disk cache. 
On-disk cache
hits are valid I/O anyway.

Brendan

-- 
Brendan Gregg, Sun Microsystems Fishworks.    http://blogs.sun.com/brendan

On Tue, Aug 11, 2009 at 07:01:19PM -0400, Vishal Ahuja
wrote:
>    Hi Brendan,
>    Thank you for the reply. I was referring to the file system cache.
>    Sorry, I am not aware of io provider?
Ok; if you are interested in disk service time, then you can trace from
the disk driver or abstraction.  The io provider does this:

	http://wikis.sun.com/display/DTrace/io+Provider

Brendan
>    Thanks,
>    Vishal
> 
>    On Tue, Aug 11, 2009 at 6:47 PM, Brendan Gregg - Sun Microsystems
>    <[1]brendan at sun.com> wrote:
> 
>      G''Day Vish,
>      On Tue, Aug 11, 2009 at 03:39:01PM -0700, vhiz wrote:
>      > Hi,
>      >
>      > I am trying to figure out the service time distribution for disk
>      reads using dtrace. I am using Solaris Community edition, with UFS
>      as the filesystem. I have written simple scripts which use fbt, but
>      am not getting any consistency in my results. The biggest hindrance
>      seems to be the caching effects. Any suggestions on how to get an
>      average even though caching effects are there?
>      Which cache do you mean? If you use the io provider, you should be
>      close enough
>      to disk access that you''ll only be affected by the on-disk
cache.
>      On-disk cache
>      hits are valid I/O anyway.
>      Brendan
>      --
>      Brendan Gregg, Sun Microsystems Fishworks.
>      [2]http://blogs.sun.com/brendan
> 
> References
> 
>    1. mailto:brendan at sun.com
>    2. http://blogs.sun.com/brendan
-- 
Brendan Gregg, Sun Microsystems Fishworks.    http://blogs.sun.com/brendan