Spotted this via slashdot.org:
http://linuxgazette.net/102/piszcz.html
Thought I give this a mention.
C
On Tuesday 11 May 2004 21:28, Craig wrote:
Spotted this via slashdot.org:
Interesting, but hardly conclusive and the test doesn't take into account how different filesystems scale differently on higher/lower performance hardware. This simply confirms what I had suspected for a long time, the choice of filesystem depends on your specific needs.
Also I was a bit dismayed at the choice of image file format for the graphs, is it my eyes or is the text on most of those damm near impossible to read. PNG or GIF would have surely been a better choice (and probably resulted in a smaller file too)
On Wed, 2004-05-12 at 00:45, Wayne Stallwood wrote:
On Tuesday 11 May 2004 21:28, Craig wrote:
Spotted this via slashdot.org:
Interesting, but hardly conclusive and the test doesn't take into account how different filesystems scale differently on higher/lower performance hardware. This simply confirms what I had suspected for a long time, the choice of filesystem depends on your specific needs.
It will make me think about what filesystem i use next time i install Linux. I can't see it happening anytime soon thanks to Debian's easy upgrade, maybe a harddisk failure would do the job.
Also I was a bit dismayed at the choice of image file format for the graphs, is it my eyes or is the text on most of those damm near impossible to read. PNG or GIF would have surely been a better choice (and probably resulted in a smaller file too)
The guy had been slashdoted, id use the low quality/low bandwidth image format as well ;). I found it very hard to read as well though, i think scalable vector graphics would be the best choice for any graph images. It would surly only be a few lines of text for each graph and would keep its image quality at higher resolutions. I don't know if browsers support the format yet but Mozilla does.
Dennis
On 2004-05-12 03:50:27 +0100 Dennis Dryden ddryden@ntlworld.com wrote:
On Wed, 2004-05-12 at 00:45, Wayne Stallwood wrote:
[...] PNG or GIF would have surely been a better choice (and probably resulted in a smaller file too)
^^^^^^^^^^^
The guy had been slashdoted, id use the low quality/low bandwidth image format as well ;). [...]
Large blocks of solid colour should compress fairly well with PNG's zlib-style compression, maybe even better than jpg because of the sharp edges. I think Wayne probably had it right.
On Wed, 2004-05-12 at 00:45, Wayne Stallwood wrote:
Interesting, but hardly conclusive and the test doesn't take into account how different filesystems scale differently on higher/lower performance hardware. This simply confirms what I had suspected for a long time, the choice of filesystem depends on your specific needs.
In the generic use-case (I.E. a Debian system) I recommend XFS. I used to recommend ReiserFS, but XFS is simply a more mature solution. It lacks one or two of the bells & whistles of ReiserFS (E.g. tail packing) but overall it's a more solid option.
D. (I run XFS on my servers in telehouse; my desktop at work; my laptop; blahblahblah :-)
Daniel Silverstone wrote:
In the generic use-case (I.E. a Debian system) I recommend XFS. I used to recommend ReiserFS, but XFS is simply a more mature solution. It lacks one or two of the bells & whistles of ReiserFS (E.g. tail packing) but overall it's a more solid option.
Indeed - we run XFS exclusively at The Moving Picture Company. All render farm, file servers and workstations are running XFS and it's journaling capabilities have saved us many a time.
Regards,
Martyn
Martyn Drake wrote:
Daniel Silverstone wrote:
In the generic use-case (I.E. a Debian system) I recommend XFS. I used to recommend ReiserFS, but XFS is simply a more mature solution. It lacks one or two of the bells & whistles of ReiserFS (E.g. tail packing) but overall it's a more solid option.
Indeed - we run XFS exclusively at The Moving Picture Company. All render farm, file servers and workstations are running XFS and it's journaling capabilities have saved us many a time.
We use ext3: last I heard XFS had a few problems. AOK now?
Cheers, Laurie.
On Wed, 2004-05-12 at 09:52, Laurie Brown wrote:
We use ext3: last I heard XFS had a few problems. AOK now?
I've had nought but problems from ext3, and nought but easiness from ReiserFS and XFS.
D.
On 2004-05-12 09:51:44 +0100 Daniel Silverstone dsilvers@digital-scurf.org wrote:
I've had nought but problems from ext3, and nought but easiness from ReiserFS and XFS.
Counter-point: The ReiserFS systems here have caused enough problems (corrupt files :-( )that they will go away at next upgrade; Hans Reiser is also currently really unhappy at his Debian packaging and screwing around with the licence. ext3 hasn't caused problems here apart from having to unmount to resize (IIRC... it's not a frequent operation).
Not used XFS on a real live system, but I remember someone inadvertantly converting their /usr to XFS while leaving the XFS tools on it, so take care when converting, of course...
On Wed, May 12, 2004 at 10:25:20AM +0100, MJ Ray wrote:
On 2004-05-12 09:51:44 +0100 Daniel Silverstone dsilvers@digital-scurf.org wrote:
I've had nought but problems from ext3, and nought but easiness from ReiserFS and XFS.
Counter-point: The ReiserFS systems here have caused enough problems (corrupt files :-( )that they will go away at next upgrade; Hans Reiser is also currently really unhappy at his Debian packaging and
I have had ReiserFS problems in the past which resulted in partitions going *boom* and melting so I trust it about the same as FAT32...
screwing around with the licence. ext3 hasn't caused problems here apart from having to unmount to resize (IIRC... it's not a frequent operation).
I did once have a large ext3 partition suddenly lose all of its contents which was rather worrying as the 100GB of data was not backed up at the time (wasn't my fault, the user had not put the data onto the network as they were in a hurry) although a reboot and a fsck recovered the data we were *very* alarmed at the time as we had 24 hours before the deadline at that point and we just lost 5 days of work... I think that was on kernel 2.4.23 which ext3 was still marked experimental on but the subsequent release squashed that bug and the driver was no longer marked experimental.
Not used XFS on a real live system, but I remember someone inadvertantly converting their /usr to XFS while leaving the XFS tools on it, so take care when converting, of course...
When I first tried XFS it had massive problems related to breaking everything totally but that was a loooong time ago when it had big warnings on it. From what I have read it has come a long way since then so maybe I will try it on the laptop and new machines around here and see what happens.
Adam
adam@thebowery.co.uk wrote:
When I first tried XFS it had massive problems related to breaking everything totally but that was a loooong time ago when it had big warnings on it. From what I have read it has come a long way since then so maybe I will try it on the laptop and new machines around here and see what happens.
I think I'll set it up on a machine to mirror one of our backup servers, and see how that goes too.
Cheers, Laurie.
Laurie Brown wrote:
We use ext3: last I heard XFS had a few problems. AOK now?
Been running on over 600 render farm boxes, 300 workstations, and 20 odd fileservers supporting a total (roughly) of 21Tb of online storage for the past several years. I should think that qualifies as being okay :)
Regards,
Martyn
On 12 May 2004, at 10:03, Martyn Drake wrote:
Laurie Brown wrote:
We use ext3: last I heard XFS had a few problems. AOK now?
Been running on over 600 render farm boxes, 300 workstations, and 20 odd fileservers supporting a total (roughly) of 21Tb of online storage for the past several years. I should think that qualifies as being okay :)
Impressive.
I knew this article would get some people talking on here. Mind you, I did feel some results wasn't quite right and he was using ATA 100 IDE. Would be interesting to see how SCSI and SATA goes on there.
C
On Wed, May 12, 2004 at 10:16:07AM +0100, Craig wrote:
I knew this article would get some people talking on here. Mind you, I did feel some results wasn't quite right and he was using ATA 100 IDE. Would be interesting to see how SCSI and SATA goes on there.
It would have been better if all the filesystems had been tuned up also I think looking at that article they were all left in their default state.
Another thing I would liked to have seen was perhaps benchmarks related to apache, NFS, sql and Samba performance (etc.) when using these filesystems rather than just raw disk operations as usually once you have the data off the disk there is subsequently a network operation involved.
Adam
On 12 May 2004, at 10:34, adam@thebowery.co.uk wrote:
I knew this article would get some people talking on here. Mind you, I did feel some results wasn't quite right and he was using ATA 100 IDE. Would be interesting to see how SCSI and SATA goes on there.
It would have been better if all the filesystems had been tuned up also I think looking at that article they were all left in their default state.
Default vs tuned would be ideal.
Another thing I would liked to have seen was perhaps benchmarks related to apache, NFS, sql and Samba performance (etc.) when using these filesystems rather than just raw disk operations as usually once you have the data off the disk there is subsequently a network operation involved.
Now that's something I like to see, I would like to see some real world testing as well (but how to construct that is going to be difficult).
C
On Wed, May 12, 2004 at 09:52:33AM +0100, Laurie Brown wrote:
Martyn Drake wrote:
Daniel Silverstone wrote:
In the generic use-case (I.E. a Debian system) I recommend XFS. I used to recommend ReiserFS, but XFS is simply a more mature solution. It lacks one or two of the bells & whistles of ReiserFS (E.g. tail packing) but overall it's a more solid option.
Indeed - we run XFS exclusively at The Moving Picture Company. All render farm, file servers and workstations are running XFS and it's journaling capabilities have saved us many a time.
We use ext3: last I heard XFS had a few problems. AOK now?
I'm an ext3 user too; I reckon it's probably had more testing than most of the others and I haven't seen any problems with it. XFS only recently got added to mainline 2.4 kernels, didn't it? Might give it a look next time I'm building a box I can play about with.
J.
On 12 May 2004, at 11:13, Jonathan McDowell wrote:
I'm an ext3 user too; I reckon it's probably had more testing than most of the others and I haven't seen any problems with it. XFS only recently got added to mainline 2.4 kernels, didn't it? Might give it a look next time I'm building a box I can play about with.
You might as well shift to 2.6 kernels. I have installed those throughout this department and they have been rock solid on desktop and servers.
C
On Wed, May 12, 2004 at 01:17:58PM +0100, Craig wrote:
On 12 May 2004, at 11:13, Jonathan McDowell wrote:
I'm an ext3 user too; I reckon it's probably had more testing than most of the others and I haven't seen any problems with it. XFS only recently got added to mainline 2.4 kernels, didn't it? Might give it a look next time I'm building a box I can play about with.
You might as well shift to 2.6 kernels. I have installed those throughout this department and they have been rock solid on desktop and servers.
I'm running 2.6 on my laptop and a couple of local test boxes, but the majority of machines I run are in colo and as such I'd rather stick with 2.4 until 2.6 matures a bit more. Though probably by the time I'm ready to run XFS on any production box I'll consider 2.6 production ready.
J.
On 2004-05-12 00:45:10 +0100 Wayne Stallwood ALUGlist@digimatic.plus.com wrote:
Interesting, but hardly conclusive and the test doesn't take into account how different filesystems scale differently on higher/lower performance hardware.
I was surprised that there was only a "total time" used to make the conclusion and no weighted average based on relative frequency of the different acts (remove/copy/write and so on) tested.
This simply confirms what I had suspected for a long time, the choice of filesystem depends on your specific needs.
It will always be so, I suspect.