[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: fsfs and Linux filesystems

From: Stefan Urbat <stefan.urbat_at_apastron.lb.shuttle.de>
Date: 2005-08-29 05:41:13 CEST

On Sun, 28 Aug 2005, Christopher Ness wrote:

> On Sun, 2005-08-28 at 11:01 +0200, Bernd Rinn wrote:
>> Since fsfs is build on top of the native filesystem without a middle
>> layer it should be more dependent on the performance characteristics of
>> the native filesystem than bdb.
>>
>> Does anyone have experience with which Linux filesystem works best with
>> fsfs (ext3, reiserfs, XFS, ...)? Or maybe which one to avoid? Or which
>> mount options to set in order to get good performance?
>>
>> Any piece of advice would be appreciated.

> Therefore this article about FS suggests that ReiserFS should be the
> best for storing many small files.
> http://www.linux-mag.com/2002-10/jfs_01.html
>
> To quote the article, about 1/2 way down the page:
> "ReiserFS is about eight to fifteen times faster than Ext2 at handling
> files smaller than 1K."
>
This FS may be best in matters of performance, but is the worst at now
regarding data safety: you should avoid ReiserFS IMHO under all
circumstances for truely important data on truely important servers. I had
several incidents with it in the past; it is disintegrating itself rather
quickly under not optimal circumstances.

My advice: use xfs, because it is the best balance at now in terms of
safety AND performance considered together. And there are no version
differences/mismatches, there is just ONE xfs version, while ReiserFS is
as bad as NTFS, incompatible in several versions of it (3.5, 3.6 and 4 are
all completely incompatible in one direction: you can't use ever 4 on a
3.6 aware system (all 2.4.x and so far all 2.6.x Kernels and never 3.6 on
3.5 aware systems running 2.2.x kernels with Reiser patch for example) and
you will lose performance if running an older version with a newer
kernel).

> That being said, even a simple revision was 1.3K in a FSFS type file
> system (ext3) so perhaps this is not applicable. I would expect
> revisions to be anywhere between 2K to 100K on average for text based
> revisions. Binaries and all bets are off.
>
> All that is needed is some actual data tests on the different FS'. ;)
>
> ReiserFS publishes some of their own benchmarks:
> http://www.namesys.com/benchmarks.html#mongo.2.6.11
>
They are often tweaked to make it appear better compared with others, so
these numbers should be taken with some care. For example xfs is clearly
faster when deleting big files.

Regards, Stefan Urbat

--
Registered GNU/Linux-User #169380 on http://counter.li.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Mon Aug 29 05:43:09 2005

This is an archived mail posted to the Subversion Users mailing list.