[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: benchmarks?



Hi,

On Thu, Nov 01, 2001 at 01:01:48PM -0600, Bruce Guenter wrote:
> On Wed, Oct 31, 2001 at 09:44:39PM -0500, John Ruttenberg wrote:
> > Are there any benchmark resultss that meaningfully compare performance of ext2
> > vs ext3 vs rieserfs vs sgi xfs?  What about effect on cpu usage?
> 
> I have some numbers comparing ext2, ext3, and reiserfs, but nothing on
> XFS.  What I have found is that benchmark results are *highly*
> application specific.  What works well for databases is different than
> what works well for mail servers, which is different than what works
> well for file servers.  So, then, what is your application?

In addition, CPU usage, latency and throughput are all compromises you
have to take.  So is data integrity, and so is the ability to survive
VM pressure, and so is the delay you're willing to put up with between
saving data and it hitting disk.  

Then there's the fact that a fs which works really well for benchmarks
on a cleanly created new partition may become much worse over time due
to fragmentation.

Benchmarks are useful tools, but once you've seen more than one or two
you rapidly learn not to trust them.  Keeping all the data in memory
and never writing to disk, for example, is a filesystem strategy which
gives the best benchmarks for dbench, but which is unlikely to be what
the user wants in real life.  :-)

Cheers,
 Stephen





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]