[sclug] Re: performance puzzle

[Tom Dawes-Gamble]

On Sun, 2005-07-10 at 14:14 +0100, Graham wrote:
> A HOLE! (Thats the file, not you ...)
> 

Hmm.... explain more.

> You are misdirecting us into presuming that you did
> what we would, in such a circumstance. (perf testing)
> This is a trick question.
> 

Well that depends on your point of view IMVHO.  I prefer to think of it
as a puzzle not a trick question.  For sure your working with incomplete
information.  But then how often are we able to work with complete
information in a performance testing/tuning environment?  

> > 4763dafae2d293e98b62979a00308158  baz
> > 
> > real    0m21.172s
> > user    0m3.010s
> > sys     0m1.225s
> > 
> > 4763dafae2d293e98b62979a00308158  foo
> > 
> > real    0m4.554s
> > user    0m3.201s
> > sys     0m1.341s
> > # 
> 
> The CPU still has to do similar work, (more!)
> but the disk transfer isnt "sand through the hourglass",
> saving 15-21 seconds.

> The interesting thing is that the user + sys is higher.
> Can you run this many times and get an average?
> Maybe the sampling whilst blocked is inaccurate.
> What happens with time cat or time dd ? (md5 is cpu bound)

Yes for some reason foo does burn more 'user' and 'sys' time than baz.
time cat takes for ever due to the way it does it's i/o.  dd give better
times so to speak.

# time dd if=/mnt/tmp/foo of=/dev/null bs=1024
1048577+0 records in
1048577+0 records out

real    0m3.143s
user    0m0.466s
sys     0m2.665s

# time dd if=/mnt/tmp/bar of=/dev/null bs=1024
1048577+0 records in
1048577+0 records out

real    0m21.196s
user    0m0.401s
sys     0m2.319s

Are typical timings.  the differs on a dd is only 0.05s 

regards,
tom.


regards,
tom.

-- 
There are 10 types of people in the world.
Those that understand Binary and those that don't.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.tmdg.co.uk/pipermail/sclug/attachments/20050710/6601a9ff/attachment.bin