Hi Sanjeev
linking the application to the ARCSTAT_BUMP(arcstat_hits) is not
straightforward and time consuming especially if I am running many
experiments.
Brendan has commented on on the post by providing an alternative DTrace
script , here is his post ...
------------------
dtrace -n 'sdt:::arc-hit { @[execname] = count(); }'
Or to filter just on firefox:
dtrace -n 'sdt:::arc-hit /execname == "firefox-bin"/ { @ = count(); }'
If neither of them work, it just means that that particular sdt provider
probe wasn't available, and we can get this information from using the fbt
provider instead.
Note: ZFS prefetch is pretty effective, so streaming reads from disk will
often
show up as arc hits since ZFS has put the data in cache before the
application
reads it. Which is the intent.
------------------------
I ran Brendan script while watching arcstat from kstat and notice great deal
of discrepancy in terms of cache hits & misses numbers between kstat &
Brendan's scrip. I am not sure exactly where 'sdt' probe collects the cache
hits information !!.
Any feed back ?
On Wed, Feb 10, 2010 at 11:43 PM, Sanjeev <sanjeev.bagew...@sun.com> wrote:
> Abdullah,
>
> On Tue, Feb 09, 2010 at 02:12:24PM -0500, Abdullah Al-Dahlawi wrote:
> > Greeting ALL
> >
> > I am wondering if it is possible to monitor the ZFS ARC cache hits using
> > DTRACE. In orher words, would be possible to know how many ARC cache hits
> > have been resulted by a particular application such as firefox ??
>
> ZFS has kstat for a whole bunch of these. You could run "kstat -m zfs" and
> see
> if which ones help you.
>
> Correlating these stats to a particular application would be a little
> difficult.
>
> Using Dtrace, you could track down the read/write operations and see if
> they
> lead to a hit/miss in ARC. Take a look at the routine arc_read here :
>
> http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/fs/zfs/arc.c
>
> "ARCSTAT_BUMP(arcstat_hits);" is the routine which increments the hit
> count.
>
> You could track arc_read()s called from the context of the application and
> check
> if it causes a hit or a miss.
>
> Thanks and regards,
> Sanjeev
> >
> > Your response is highly appreciated.
> >
> >
> > Thanks
> >
> >
> > --
> > Abdullah
> > dahl...@ieee.org
> > (IM) ieee2...@hotmail.com
> > ----
> > Check The Fastest 500 Super Computers Worldwide
> > http://www.top500.org/list/2009/11/100
>
>
> --
> ----------------
> Sanjeev Bagewadi
> Solaris RPE
> Bangalore, India
>
--
Abdullah Al-Dahlawi
http://www.google.com/profiles/aldahlawi
----
Check The Fastest 500 Super Computers Worldwide
http://www.top500.org/list/2009/11/100
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
