Re: some linux tuning (fs benchmarks)

Doug Renner writes:
[jlewis writes:]
> > bonnie runs that fit entirely in memory are virtually meaningless as
> > benchmarks.
>
> I don't share that feeling. I'm sure they're useless when you are trying
> to extrapolate specific data (I.E. like pure FS speed), but when all that
> is important is how fast a given environment can cram stuff on the disk,
> it's good for me.

  You're missing his point. When you run Bonnie with 1MB, 10MB, and
100MB files, you haven't measured any aspect of how fast your system
can "cram stuff on the disk." I went and checked the Bonnie reference
site to be sure of this, because I'm also interested in the file/disk
system benchmarking issue. The Bonnie benchmark does operations on a
*single* file. If that file fits entirely in your system's RAM, all
you've measured is how fast you can move blocks of data in and out of
your file cache in RAM. There's never any reason to run Bonnie on 1MB
files unless your OS does no disk caching at all (get a new one!) or
you have less than 1MB total RAM in your machine (the good ol' days...)

  In short, you weren't measuring what you thought.

  At a bare minimum, if you have 256MB RAM, you should be testing on
something like a 512MB file to get any meaningful results at all out of
"Bonnie". From the author:

"Bonnie performs a series of tests on a file of known size. If the size
is not specified, Bonnie uses 100 Mb; but that probably isn't enough
for a big modern server - you (need) your file to be a lot bigger than
the available RAM." <http://www.textuality.com/bonnie/intro.html>

  However, testing Bonnie on a single 512MB file is also not going to
relate very well to your performance on Squid, because for Web caching,
you need to be able to create and access lots of small files very fast
in random order, so it's really issues like a particular file system's
performance on large-scale file creates, directory updates, etc. that
will dominate. That's why Alex Rousskov said the benchmark needs to
fit Squid's usage patterns (though I still think it doesn't need to be
specific to Squid - lots of apps have at least roughly that kind of
macro usage pattern.)

  -- Clifton

-- 
 Clifton Royston  --  LavaNet Systems Architect --  cliftonr@lava.net
        "An absolute monarch would be absolutely wise and good.  
           But no man is strong enough to have no interest.  
             Therefore the best king would be Pure Chance.  
              It is Pure Chance that rules the Universe; 
          therefore, and only therefore, life is good." - AC