Message-ID: <BAY101-F22C68A84110DBB2CA82199D0B60@phx.gbl>
In-Reply-To: <16259.1124725300@sss.pgh.pa.us>
From: "Jignesh Shah" <jigneshk@hotmail.com>
To: tgl@sss.pgh.pa.us
Cc: pgsql-performance@postgresql.org
Subject: Re: MemoryContextSwitchTo during table scan?
Date: Mon, 22 Aug 2005 18:37:21 -0400
Mime-Version: 1.0
Content-Type: text/plain; format=flowed


Hi Tom,

Like I mentioned I am using DTrace on Solaris 10 x64 and not gprof.
DTrace is not based on sampling but actual entry/exit point. Ofcourse my 10 
second profile is just a sample that I can assure you is representative of 
the query since it is a very simple query that does simple table scan. (I am 
taken profiles at different times of the queries all giving similar outputs)

In case of DTrace I am actually measuring "wall clock" for leaf functions.

For more information on DTrace please refer to:
http://docs.sun.com/app/docs/doc/817-6223/6mlkidlf1?a=view

Regards,
Jignesh


----Original Message Follows----
From: Tom Lane <tgl@sss.pgh.pa.us>
To: "Jignesh Shah" <jigneshk@hotmail.com>
CC: pgsql-performance@postgresql.org
Subject: Re: [PERFORM] MemoryContextSwitchTo during table scan?
Date: Mon, 22 Aug 2005 11:41:40 -0400

"Jignesh Shah" <jigneshk@hotmail.com> writes:
 > Running a script (available on my blog) I find the following top 5 
functions
 > where it spends most time during a 10 second run of the script

It's pretty risky to draw conclusions from only 10 seconds' worth of
gprof data --- that's only 1000 samples total at the common sampling
rate of 100/sec.  If there's one function eating 90% of the runtime,
you'll find out, but you don't have enough data to believe that you
know what is happening with resolution of a percent or so.  I generally
try to accumulate several minutes worth of CPU time in a gprof run.

 > MemoryContextSwitchTo and LockBuffer itself takes 15% of the total time 
of
 > the query. I was expecting "read" to be the slowest part (biggest 
component)
 > but it was way down in the 0.4% level.

You do know that gprof counts only CPU time, and only user-space CPU
time at that?  read() isn't going to show up at all.  It's fairly likely
that your test case is I/O bound and that worrying about CPU efficiency
for it is a waste of time anyway.

			regards, tom lane