On Fri, 2004-05-21 at 18:32, Kaj Haulrich wrote:

> > By way of confirmation, please witness these latest benchmarks in
> > Linux Gazette which bear out what you are saying:
> >
> > http://linuxgazette.net/102/piszcz.html
> >
> >
> > LX
> 
> Interesting. And I think the defragging can be set up as a cron job. 
> My experience with JFS stems from my OS/2 days, and I don't 
> remember I ever had to defrag it. In OS/2 there was a nifty 
> feature, a graphical representation of the filesystem. One could 
> actually see how it worked, distributing files in a very clever 
> manner. Fragmentation never amounted to more then 2 %.
> 
> Next time I do a clean install, I'll try it on linux, although I 
> have no complaints whatsoever with my ReiserFS.
> 
> Kaj Haulrich.

<grin>  You might if you look at the benchmarks.  Reiser consumes a
horde of cpu cycles in order to deliver performance that is not really
superior either to xfs or jfs *overall*.  I tip the hat to Reiser
design, it's quite sophisticated, however if the design is not living up
to it's promise, then I'm unsure of it's true value.

Look at image034.jpg (Split 10mb File) of the linux gazette article
page.  Reiser's advanced design should be letting it kick ass
here...yet, it is just barely comparable to  JFS's curve, which is
phenonmenal.

In image029.jpg (Megabytes Per Second On Copy From Current To Other
Disk), Reiser has the poorest showing in the group, with JFS beating it
out.  Which brings me to the topic of image021.jpg -- "Total CPU
Utilization For Each Filesystem".  This test was an aggregate
accumulation of all cpu usage across all tests.  JFS shows an
unbelievable kick ass advantage over the other journaling filesystems,
and then in addition shows an advantage over ext2, which isn't even a
journaling filesystem, for pete's sake!!!

That's an absolute first; it's never been done before.  I do not know
why it has not gotten more press.  Logic would lead you to believe that
such a feat as that would not be possible. Journaling filesystems have
to do more stuff than nonjournaling filesystems. But yet, there it is; a
journaling filesystem is outperforming a non journaling filesystem in
CPU utilization benchmarks.

Reiser, on the other hand, has the absolute worst showing of any of the
other filesystems here in image021.jpg.  Of the entire group, Reiser is
the biggest CPU hog; while at the same time delivering no overall speed
advantages to it's CPU usage.  The *aggregate* tests reveal this.

Given the other aggregate benchmarks (like image020.jpg, Total Time For
Each Filesystem), the final qualifier, for me anyway, becomes the
advantage of total system speed.  JFS has the undisputed advantage here;
it leaves the CPU available for things other than testing sophisticated
filesystem theory.  ;)

LX


____________________________________________________
Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com
Join the Club : http://www.mandrakeclub.com
____________________________________________________

Reply via email to