On Thu, 6 Jan 2005 14:07:05 -0800 (PST)
death rince <[EMAIL PROTECTED]> wrote:

> Hi,
> 
> I was browsing through benchmarks of Reiserfs and
> other filesystem when I came across this,
> 
> http://linuxgazette.net/102/piszcz.html
> 
> Can anyone please give an insight, behind the
> rationale of such tests and what exactly it signifies
> in terms of working of the kernel with respect to
> different filesystems.

IMHO this benchmark is rather pointless (and faulty).
A sync between each run is not enough. A reboot would be better. Also, the
time for "sync" should be included in the timing, IMO.

Additionally this page doesn't mention mount options.
For example, the time required to open a mail folder containing ~2000
messages in Sylpheed dropped from ~8 seconds to <1 second when using the
"notail" option in reiserfs. This disables efficient storage of very small
files (<4kB) and "tails" of larger files. In Linux, this feature is unique
to ReiserFS anyway (NTFS in Windows supports this as well).

If you want to find the best all-purpose filesystem use a proven benchmark
like bonnie, bonnie++ or tiobench.
If you have a specific application, just benchmark the real thing.

If you're running an average desktop system, simply choose ext3, xfs or
reiserfs(with "notail"). The quality of the kernel's IO-scheduling will
outweight any filesystem differences anyway.

If you are doing video *recording* (especially multiple streams at once)
choose XFS. All other filesystems suffer enormously when doing
simultaneous streaming writes.

Regards

--
[email protected] mailing list

Reply via email to