Journaling Filesystems (Re: tlug: Ext3)

["Oliver M . Bolzer" <oliver@example.com>]

On Mon, Mar 06, 2000 at 11:23:35PM +0900
Austin Kurahone wrote...

> I've been meaning to post this for a while (Ok, at least since the last
> nomikai), butI've been too busy.a  Since a bunch of people seemed
> interested in tryng out ext3, here's my quick and dirty guide
> (<PLUG>The april issue of Linux Japan also has a article on this by
> yours truely</PLUG>)

I have done some testing on the subject ext3 VS Reiserfs
(<AOL TYPE="metoo">Detailed benchmarks and results in Linux Japan May</AOL>)
Ext3 is really not yet where we would like it. The drawback is that all
write operations double in time, since all data has to be written
twich due to full-journaling.
If you need journaling, use ReiserFS, is is not only more stable, but
is also as fast or sometimes even faster then ext2. The problem is that
you can't boot from it, so you'll need a separate /boot partition.
The only place you want to use ext3 is, for a pretty limited / which
is not offen written to or a fairly static data partition (like MP3s)
for which you don't have the extra space to switch from ext2 to ReiserFS.
The only advantage ext3 offers is not to need to reformat.
If anybody need's an ext3 patch for 2.2.14, mine can be found at
http://www.sukisuki.org/linux/ I really hope it's seemingly stopped
development gets moving again real soon.

I've also attached some benchmarks results. I won't explain what each benchmark
did, but the test have been done on a DualCeleron 300 with 128MB RAM and
8GB partitions and a PIII 500MHz x 4 with 2GB system with a 140GB RAID-5
system.

write-tests          magi                            |   edelweiss
		ext2	ext3   reiserfs			ext2	reiserfs
maketestdir	 6:18	12:39	 5:43	1: 2  : 0,9	1:16	1:29
maketestdir x4	22:55	45:26	20:25	1: 1,9: 0,9	5:50	5:10
dd 1000MB	 0:18    1:20	 0:21	1: 4,4: 1,1	-	-
append		 2:12	 3:24	 3:46	1: 1,5: 1,7	1:20	1:36

read-tests
		ext2	ext3  reiserfs			
cat 100MB	 0:29	 0:29	 0:27	1: 1  : 0,9	 -	-
readrnd		 0:11	 0:13	 0:12	1: 1,2: 1,1	0:05	0:04
	
Also if you want to learn kernel internals and implementation of a
filesystem, watch IBM's JFS. When first released, it could only mount
and ls (showing . adn ..). After 4 weeks touch,mkdir,rm,rmdir work.
If things wo in this speed, we'll be able to read and write to files
in 4 weeks.

Regarding SGI's XFS, I have hearded that they have it working,
but it has sooooo bad performance, they can't show it publicly.

Filesystems are a hot topic to watch out, this year as 
all these evolve further.


-- 

	Oliver M. Bolzer
	oliver@example.com

$B!V?x$(A7?)$o$L$b!"CK$N@example.com$-MM!W(B

GPG (PGP) Fingerprint = 621B 52F6 2AC1 36DB 8761  018F 8786 87AD EF50 D1FF
--------------------------------------------------------------------
Next Technical Meeting: March 11 (Sat) 13:00 Temple University Japan
* Topic: "What's new in Perl 5.6"
Guest speaker: Simon Cozens (TLUG Perl guru)
Next Nomikai Meeting: April 20 (Thu) Linux Conference 2000 Spring Ed.
--------------------------------------------------------------------
more info: http://www.tlug.gr.jp        Sponsor: Global Online Japan