The Filesystem choice thread

Message

Lovechild · Post by **Lovechild** » Thu Nov 07, 2002 1:16 pm

I'm surprised that noone has mentioned JFS yet, I used this FS for a while on Mandrake, and while a bit unstable it seems fast.

I'm personally looking forward Reiser4, I currently use ReiserFS3 and it serves me well, good speed and good stability, but the new release is said to be 50-100% faster.... now that's quite a chunk of cheeze....

pilla · Post by **pilla** » Thu Nov 07, 2002 2:05 pm

My sense of adventure is far away from my thesis

boyo wrote:C'mon, Where's your sense of adventure? ext3 has done well when I've used it, but that hasn't been as extensively as Reiserfs or XFS (both have been outstanding). Being the performance junkie I am if there's something faster that's not ridiculously unsafe I'm all about it. The mounting without a journal sounds handy though.

dpowers · Post by **dpowers** » Thu Nov 07, 2002 5:14 pm

This is an excellent thread, and Reiser appears to be the general winner. Besides the -tail/-notail stuff, what are some other things that one should know about Reiser. Any other flags to set? Any other useful tidbits of knowledge about it?

Thanks for all the great great info.

-dpowers

squanto · Post by **squanto** » Thu Nov 07, 2002 5:32 pm

dpowers wrote:This is an excellent thread, and Reiser appears to be the general winner. Besides the -tail/-notail stuff, what are some other things that one should know about Reiser. Any other flags to set? Any other useful tidbits of knowledge about it?

Thanks for all the great great info.

-dpowers

Check out http://www.namesys.com/ they are the developers. And I am sure they know more than I do

enjoy

boyo · Post by **boyo** » Thu Nov 07, 2002 7:52 pm

lupad, what a great post. Thanks fo rthe info. You're the man. Is your controller ATA-133?

As is stands Reiser4 looks to be the hands down winner and their website has Dec. 31 listed as the due date. Whether that's for the FS itself or just a finalized revision of that document I have no idea. I'd like to assume it's the release date for Reiser4. I'm wondering if the SGI camp is going to cook up anything big in the near future. It's been a while since I've heard a peep from them about XFS.

mooman · Post by **mooman** » Fri Nov 08, 2002 1:34 am

I'm a noob on the FS subject, but wanted to point out an application consideration for filesystems...

If reiser is superior for smaller files, then an excellent match would be a mailserver (or mail proxy, or even a personal box) that uses the maildir format. This approach stores each email as a seperate text file, so it would stand to benefit considerably from that optimization.

Ditto for a web proxy like squid. Squid expects the average file size of cached objects to be around 13K. Which means there will be quite a few 1k to 5K type files that would benefit from an optimal filesystem.

Conversely, an MP3 server or GIMP workstation probably don't gain much from small file improvements.

Anyway, just wanted to reframe the conversation from "Which one is best?" to "which one is best for the role of the box?"... Of course, stability is a whole 'nuther issue.

boyo · Post by **boyo** » Fri Nov 08, 2002 2:38 am

That's a good point. However, if Reiser4 lives up to it's own documentation then we should have a clear winner in nearly every category. Dec. 31 is just over a month away. It won't long till we find out (hopefully).

eryvile · Post by **eryvile** » Sat Nov 09, 2002 2:27 pm

boyo wrote:Honestly, hdparm stats are hardly anything to feel miserable about. The intent of my question was curiosity, not to discredit anyone. If 45-50 MB/s was mis-read, no biggie. If not give me a hand tweaking hdparm for sure.

Ok, today I finally managed to verify those numbers. And to my joy

, I wasn't that far off last time:

Code: Select all

/dev/hdb:
 Timing buffer-cache reads:   128 MB in  0.58 seconds =220.69 MB/sec
 Timing buffered disk reads:  64 MB in  1.41 seconds = 45.39 MB/sec

For teaking, honestly, I didn't do anything except for compiling all support for the VIA vt8233 into the kernel (gentoo-sources 2.4.19-r9). It was already mentioned somewhere in this thread that putting the right chipset into your kernel can already do most of the tweaking which obviously worked for me

. Here's also some of the dmesg output:

Code: Select all

VP_IDE: VIA vt8233 (rev 00) IDE UDMA100 controller on pci00:11.1
    ide0: BM-DMA at 0xfc00-0xfc07, BIOS settings: hda:DMA, hdb:DMA
    ide1: BM-DMA at 0xfc08-0xfc0f, BIOS settings: hdc:DMA, hdd:DMA
hda: IC35L040AVER07-0, ATA DISK drive
hdb: IC35L040AVVA07-0, ATA DISK drive
hdc: DVD-ROM DDU1621, ATAPI CD/DVD-ROM drive
hdd: PLEXTOR CD-R PX-W1610A, ATAPI CD/DVD-ROM drive
ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
ide1 at 0x170-0x177,0x376 on irq 15
hda: 80418240 sectors (41174 MB) w/1916KiB Cache, CHS=5005/255/63, UDMA(100)
hdb: 80418240 sectors (41174 MB) w/1863KiB Cache, CHS=5005/255/63, UDMA(100)
hdc: ATAPI 40X DVD-ROM drive, 512kB Cache, UDMA(33)
Uniform CD-ROM driver Revision: 3.12
hdd: ATAPI 40X CD-ROM CD-R/RW drive, 2048kB Cache, DMA

Also interesting to mention, the disk on hda, although it is basically the same model, is about 7 megs slower (it holds my w2k installation, that's why its performance isn't a big deal for me when using Linux).

Ok, back to the original intention of this thread, the question for the right filesystem. I already mentioned the article in the german magazin iX where they compared different journaling filesystems: BSD FFS with soft updates, ext3, reiser, jfs and xfs. As, unfortunatley, they didn't put this article online, here are some numbers

Code: Select all

                         cp -R       rm -rf
FreeBSD, soft updates     2:39        0:55
Linux, ext3               2:39        1:38
Linux, xfs                2:41        1:05
Linux, reiserfs 3.6       2:33        0:14
...
source: iX 10/2002, p. 129

All test were done with /usr/ports of FreeBSD which had at that time 101980KB in 47812 files and 10013 directories. They also had some numbers for jfs, but those were imho so far off, I didn't put them in the table above (cp- R: 16:39, rm-rf: 3:43).

As all measured times are so close together (except for rm -rf for Reiser, which seems to rock when deleting files), I guess it doesn't really matter what you choose when just looking at the performance. Tweaking hdparm or putting the right chipset dirvers into your kernel can do more here.

So it's back to requirements like data consistency and reliability, which all three (ext3, reiser, xfs) have satisfied completley for me so far

I hope I could give some help!

McManus · Post by **McManus** » Tue Nov 12, 2002 8:41 pm

I have to say, in the past I've used ReiserFS, and it was rather fast. However, I ran into corruption problems all the time, and decided it wasn't worth it...

...However, nowadays, ReiserFS seems to be the FS of choice for a lot of Gentoo developers (Robbins included), and I have tried it recently and it does seem to be stable. However, I'm still wary of corruption problems that I've had in the past. If an FS is prone to corruption whenever the VM Manager changes (which it has quite frequently in the past), then while it might not be unstable now, it could be later.

Plus, I've NEVER had any corruption with ext3, and since data integrity is more important to me, I think I'll stick with ext3 for now. But I guess the point is, figure out what you want and try it; if it doesn't work well, try something else. It's all about your own experience. (what works for someone else might not work for you, and vice-versa)

...a good example of this is when DRobbins was saying how he never had any issues with ReiserFS in SuSE, because they tweaked their kernel to properly support ReiserFS... well, it was in SuSE that I had ALL my ReiserFS problems....

TheEternalVortex · Post by **TheEternalVortex** » Wed Nov 13, 2002 8:27 am

/dev/hda:
Timing buffer-cache reads: 128 MB in 0.41 seconds =312.20 MB/sec
Timing buffered disk reads: 64 MB in 1.39 seconds = 46.04 MB/sec

Of course, this makes up for that:

/dev/hdb:
Timing buffer-cache reads: 128 MB in 0.45 seconds =284.44 MB/sec
Timing buffered disk reads: 64 MB in 4.28 seconds = 14.95 MB/sec

Gentoo is installed on /dev/hdb

.

squanto · Post by **squanto** » Wed Nov 13, 2002 4:37 pm

Well... after borking my Gentoo install on my desktop, (i somehow got rid of glibc...) I decided to just do a mkreiserfs onto that partition and start over.
I have had some wierd problems since doing this. I would alter files, such as XF86Config and then do a startx and half of what I had changed in the file would change back or be erased entirely, the first half of the file got deleted one time.
Not sure if this is a reiser thing, but going to end up zeroing out the partition and then trying the install again... arg...

I am going to stay with reiser though.

whit · Post by **whit** » Tue Nov 26, 2002 4:45 am

I've had reiser on several production servers - Web and FTP with lots of files coming and going - as well as on several desktops for well over a year. The only case of data loss or corruption was on one of the desktops when the drive developed a bad spot - obviously hardware - an the reiser recovery utility (advertised as beta) couldn't deal with the bad spot and lost the whole partition. That noted, most of the partition was still accessible before trying to use reiser tools to recover it, and I don't know that another format would have been any better. I've never had any loss in reiser due to power losses, and by comparison ext3 seems much slower in checking its logs and partitions when coming back up after a power loss. But ext3 has the advantage that ext2 can be converted to it - running ext2 for anything but maybe a boot partition is certainly less safe.

darktux · Post by **darktux** » Tue Nov 26, 2002 1:21 pm

Here's a website for you about ext3

http://tuxslare.org/index.php?secs=extended&id=392

klimg · Post by **klimg** » Tue Dec 03, 2002 4:13 pm

Well - since this is the filesystem thread.I just installed JFS on one partition with kernel 2.4.20.All other partitions are ext3.If I move a big <200mb file between ext3 partitions it goes at about 1 MB/s.From a ext3 to the JFS partition it moves at 12 MB/s.Real strange.

CoronaLover · Post by **CoronaLover** » Tue Dec 03, 2002 5:00 pm

Welll I am using kernel 2.5.50 with ext3 because ext3 got some very big speed improvments in the 2.5 tree.
the Orlov Block Allocator, HTREE, fsync speedups (wich can sometimes couse filesystem corruption on 2.4.20...), and other misc stuff.
wanted to use Reiser4 but the latest snapshot is for kernel 2.5.48 and 2.5.48 looks like a pretty bad kernel.

klimg · Post by **klimg** » Tue Dec 03, 2002 7:46 pm

Disregard my post - I did move a file from /home which is a ok partition to JFS.When I tried to ext3 I used /data as a source which is fragmented like hell.If I do it from fragmented ext3 to JFS I get about 40% higher speed.

matthead · Post by **matthead** » Thu Dec 05, 2002 1:20 am

I put together a feature comparison table, mostly for my own use. After seeing this thread, I'd like to share.

I'd love any feedback anyone would like to give.

-Matt

squanto · Post by **squanto** » Thu Dec 05, 2002 1:49 am

matthead wrote:I put together a feature comparison table

I thought nfs was independent of the file system? no?

matthead · Post by **matthead** » Thu Dec 05, 2002 2:27 am

Well, it mostly is. I remember one of the old reiserfs problems being that some NFS operation required the inode number for a file to remain constant, and reiserfs kept returning different numbers. Or something like that.

I could've sworn I read on the JFS project website that there were problems with NFS, but I cannot seem to find what I thought I saw. I'll keep looking through the docs, and change that item in the table if I can't find anything.

frogger · Post by **frogger** » Thu Dec 05, 2002 7:40 am

I've been using reiserfs for at least 1.5 years now. Even in its "unstable" days I never experienced any data loss. I've been running it on a Slack 8.0 proxy/webserver, since shortly after Slack 8 was released, and it has been rock solid. Also running it on my Gentoo workstation without problems.

I've also tried XFS and had no problems with that either, but moved away from it as I prefer to use the vanilla kernel, and XFS is one big patch.

Also tried ext3 with no problems at one point:)

So basically, either I'm lucky or they are all relatively stable.

PlatinumCursor · Post by **PlatinumCursor** » Fri Dec 06, 2002 3:30 am

Ive been using ReiserFS ever since support first came in the 2.4 kernel - Never once have I lost data or had any corruption whatsoever. I have tried ext3, and before 2.4 I used ext2, and I must say that for speed, nothing beats reiserfs. I do alot of web design, image editing, and some gaming on the side, and I can say that for these things, Ive seen a big boost with Reiserfs.

klimg · Post by **klimg** » Mon Dec 16, 2002 12:01 pm

I had a very specific problem with ext3.When I download a file with mldokey it all gets split up in 9mb pieces.When I try to write this to a CD at 32X I get constant buffer underruns and a effective speed of about 8X.I did throw a JFS filesystem on there and now the buffer underruns are gone (min. buffer fill 98%) and speed average is about 22X.Seems to be pretty fast in piecing stuff together.

absinthe · Post by **absinthe** » Tue Dec 17, 2002 9:21 pm

My filesystem can kick your filesystem's ass.

huhmz · Post by **huhmz** » Thu Dec 19, 2002 8:26 pm

I was using ext3 and one day there was a power outage, rebooted and the fs was totally screwed and fsck like the worthless bitch it is of course could do nothing.
XFS lagged my computer under very heavy diskusage, ie copying files and running SQL with lots of inserts.
ReiserFS - Now thats what im talking about, no lag, very fast, no data corruption (so far, and I have shut down unclean a number of times). Im running it with quota patches on my Debian server I keep at the university and of course at home on my Gentoo desktop. Rock stable. And I head ReiserFS 4 is going to be 100% (in some cases) faster.
And you dont waste any space because of silly cluster sizes (unless mounted with notail)

TrueDis · Post by **TrueDis** » Thu Dec 19, 2002 9:09 pm

huhmz wrote:And I head ReiserFS 4 is going to be 100% (in some cases) faster.

Sadly, Reiser4 has been delayed until June 30, 2003

Check www.namesys.com

The Filesystem choice thread

Reiser experience - almost all good

Comparison Table

Re: Comparison Table

NFS and stuff