On Fri, 03 Feb 2006 12:15:24 +1300
Volker Kuhlmann <[EMAIL PROTECTED]> wrote:

> > files).  I use JFS on my Kanotix install, seems quite nice.
> 
> What's so nice about it? I mean, what the other filesystems don't do?
NOTE: This is all about file writing. I haven't done read tests on
these filesystems.

reiserfs:
        affected by partition load (from 20%)
        handles files between 5 - 10k very badly (slowly)
        can squeeze more space if tail packing is enabled.
        metadata journaling

ext2/3:
*       flushes to disk every 5s
*       journals data and metadata as well (configurable). Corruption
still possible if data is added to a middle of a file. 
*       hugely affected by partition load(from 15%)

jfs:
*       very low cpu overhead (lowest)
*       file writing unaffected by partition load
*       flushes to disk when the IO scheduler tells it to 
(see /usr/src/linux/Documentation/filesystems/proc.txt under io vm
heading)(/proc/sys/vm/

xfs:
*       low cpu overhead (second lowest)
*       very efficient disk usage (best) (caches heaps)
*       has option to make a partition a *realtime* partition -
        specifically designed for streaming video (I can't find how to
set this up however)
*       flushes to disk based on IO scheduler and it's own
"scheduler" (see filesystems/xfs.txt)

In Summary:
        ext3 for data integrity.
        jfs for low cpu overhead
        xfs for high throughput (Nick should use this)
        reiserfs for *very* small files (<3k)

thanks,
Isaac Devine

P.S. 
To speed up performance check hdparm and make sure the you have
multisect enabled (hdparm -m xx $driv) use hdparm -i $driv and set it
the value of MaxMultiSect. Turn off disk spindown (hdparm -S 0 $driv)
and power management (hdparm -B 255 $driv). Read man hdparm first
however.

Reply via email to