Use ReiserFS its the best combination of performance and rock solid reliability!
Use ReiserFS its the best combination of performance and rock solid reliability!
This guy posting this post must work in marketing......
Anyway, I would firstly try to tweak ext3 before jumping another FS. ReiserFS may be an alternative for me. JFS looks nice but it seems that very few people take it seriously.
ASUS K53S Ubuntu 13.04 64-bit
ASUS U36jc, Intel Core i5-480M, 2.66GHz, 4Gb, nVidia Optimus Ubuntu 12.04 64-bit
HP DV6-2113sa, AMD Turion™ II M620, 4Gb, ATI Radeon™ HD 4530; ASUS 1015PX Netbook Ubuntu 12.10 64-bit
I did about 3 months work on tests of different linux file systems for a new batch of servers we were installing as we had a crash and the ext3 file system hosed the fstab file badly. (mangled it with a few other files).
It came down to XFS and JFS as the performance winners but XFS was even worse than ext3 for power failures hence we decided on JFS.
Two very important notes with JFS, though, is that it's incredibly slow for deleting large amounts of small files and it doesn't have (as far as I can tell) quota support.
So we now run the new servers using JFS as the primary file system and manage quotas via software we've built.
I don't mean to sound harsh, and I am personally no fan of ext3 and would always show it less favor than XFS, but the OP is simply rubbish.
1-- define "part", and in what context! This is simply a meaningless statement without some kind of explication.Originally Posted by axely
2-- no, XFS is not the fastest filesystem. It is extremely contestable what the fastest filesystem is, and XFS is certainly very fast, but there is no clearly-definable "fastest filesystem". For example, XFS is faster than JFS when deleting large numbers of files (something mentioned earlier), but much slower than ext3. However, when finding large numbers of files it's twice as slow as either JFS or ext3. It's about the same speed as ext3 at copying files between disks, and faster than JFS, but then when creating a large file JFS is three times faster than ext3 and significantly faster than XFS, which is still twice as fast as ext3. Additionally, JFS tends to generate about 15% less CPU overhead than XFS. For more information, refer to this set of benchmarks on Linux Gazette, or this closer-to-realworld set at Debian Administration.
3-- I have no problem with this statement, although given the extreme claims that the OP makes, benchmarks to quantitatively demonstrate the speed differential would have been handy.
I will continue to quote items which are either suspect, require quantitative validation rather than qualitative and subjective hype, or are simply wildly outrageous---
If the same type of format is being performed, this seems unlikely. Formating an ext3 partition does take significantly more time (minutes rather than seconds), but such a large difference suggests that something more is going on. Formatting a 160 GB drive in JFS takes me no more than 3 seconds, however I don't know precisely what the format process does as opposed to XFS's one, or ext3's one. Anyway, the time taken to format is once-off, and doesn't necessarily tell us anything about the actual speed of the filesytem.1. Formatting 500GB in ext3 takes around 10-15 mins. Formatting the same drive in XFS takes (no kidding) 7 seconds (if you don't believe me try it).
This falls under the wildly outrageous category. The Linux Gazette article I linked above shows that ext3 and XFS are neck-and-neck when performing copies from the current disk to another, and that even where XFS significantly outperforms ext3 copying from another disk to the current one, ext3 is still pulling almost 30 MBps (note capital B for bytes, not lowercase b for bits). According to the article the testing drive was a Western Digital 250GB ATA/100 8MB CACHE 7200RPM, and the other drive was a Maxtor 61.4GB ATA/66 2MB CACHE 5400RPM. These were both running in a Dell Optiplex GX1, which I know from experience is not a fast computer; it has a Pentium 3 500 MHz processor and the test machine they used was outfitted with 768 MB of RAM, which would no doubt be PC133. This is a significantly slower computer than the Athlon 1.8 with 2 GB of RAM used in the OP. So it's quite evident that whatever the problem is with the network, the limiting factor is most definitely not the hard disks or the filesystem, given that the maximum speed of these is at least twice as fast as the 12.5 MBps limit of a 100 Mbps switch (in actuality the HDD would probably be three to four times as fast, regardless of filesystem). The OP's claim about a "perfect" network setup aside, something more is going on here than a difference in filesystem performance.2. Network traffic was slow. My network is setup perfect (as it is what I do for a living) and I could not achieve any greater than 5Mb/s no matter what I did in ext3, and it would eventually simmer down to around 2-3 after around 1 hour of transfer. XFS....no less than 11.1Mb/s at all times, no matter what I am doing, and no degregation of signal at all. (10/100 switch).
Again, as above. Packet loss does not occur on the filesystem; if there was 13-15% packet loss reading or writing data there would be widespread corruption and you almost certainly couldn't boot the disk. However, the fact that ext3 is significantly better than XFS at maintaining data integrity does suggest that, in light of the previously-mentioned slow network speed, perhaps something, somewhere, is causing IO issues. It's possible that there is something going on which is causing significant overhead to ext3 (which tries to work around it), while it's not degrading XFS (because XFS simply ignores it, or something else is preventing it being affected). If this is the case, then it could be a recommendation for ext3 more than for XFS. Anyway, regardless of the problem, network packet loss occurs at the network interface; not the hard disk interface.3. Packet loss. After monitoring 20 250GB transfers from one machine to the next, I achieved a 13-15% rate of packet loss using ext3. Absolutely none with XFS and the tranfer took 1/8 of the time (no kidding).
There is a consistent trend here of significant underperformance when using ext3, and what seems to be quite typical performance when using XFS. Again, my guess is that something is happening in the background which is degrading ext3 performance due to how ext3 handles data integrity, while not affecting XFS.4. Using vmware-server like I do everyday for everything, including real world projects/servers, I wondered why it was so slow. The problem...ext3. I installed vmware-server on XFS and the machines are at least, and I mean at the very least, twice as fast for boot and operations.
Perhaps this is true. Again, though, a more conservative CPU overhead is certainly no recommendation in many instance, despite the OP making it sound like ext3 is doing something wrong.5. I was wondering why no matter how I niced my transfer process I could gain no more speed than if the transfer (ftp, ssh, smb, etc) would not go any faster or use anymore cpu. The answer. XFS. It took full advantage of the multithreading capability of my cpu (as it uses raw i/o access) and my cpu (amd 32bit/64) is now being used correctly, whether for transfer or normal i/o operations (copy, create, delete, etc.)
Given that I've seen ext3 and other filesystems running on similarly specced machines exactly as the OP seems to be running in XFS, I am again led to the conclusion that something is wrong with his setup which is causing ext3 performance to degrade, while XFS is simply running as normal. On a properly configured and similarly specced system I would be extremely skeptical that anyone could guess what filesystem Ubuntu was using simply by interacting with the desktop and working in different applications. The difference in speed is just not that significant.6. Multitasking. I can honestly say, I am on a complete different level with XFS. It will make use of your processor and filesystem like you never thought possible. Nothing seems to slow down no my Seagate 7200 rpm 250Gb sata (8mb cache). It is like new I tell you.
Again we have a completely non-filesystem-related performance issue cropping up. Since Doom3 is only using the memory subsystem and GPU when playing, the fact that it is so much slower when running ext3 points to the same diagnosis as I suggested for the network issue: something is causing a significant IO bottleneck. If it is ext3 then this is clearly an abberation specific to the OP's system. Perhaps he could file a bug report.7. Gaming. Again, not even close with ext3. XFS makes doom3 and anything else run at least 75-100% better.
More context is required to make any use of this statement. The current usage of the hard disk, the number of files, their average size, etc, all contribute to the available measured space. Various filesystems pack files in various ways. Quite possibly as the hard disk fills up, XFS will use up progressively more space than ext3 would. However, I concede that XFS is better at allocating space than ext3 is.8. XFS gave me 6 gig back on my harddrive (98.02% usage).
This indicates nothing except that ext3 has more conservative fsck settings, while XFS is frankly reckless. There are a wealth of benchmarks and articles available online which will prove quite conclusively that ext3 is far better at maintaining data integrity than XFS is; indeed, XFS is generally kept off production machines for the exact reason that filesystem corruption is common when using it. Judging stability by fsck settings merely belies significant ignorance on the part of the OP, and throws serious doubt over all his findings.9. Stability. Well, I hate to say it, but XFS is more stable than ext3. I did approximately 15 cold shutdowns while in the middle of processing for both filesystems. Had to do 3 checks with ext3. Zero with XFS
Again, this points to an IO problem in ext3. Since the difference between the 2 GB of RAM that the OP uses, and the 4 GB he claims to feel like he has, is entirely indistinguishable for normal desktop use, this is not even subjective opinion; it's merely nonsense. Again, the OP displays ignorance both of the difference that hardware will make to the user environment, and of the way in which that hardware works. His implication that having 2 GB of swap and 2 GB of RAM should make the computer "feel" like it has 4 GB of RAM is simply absurd, since RAM not only has a hugely larger bandwidth than a hard disk, but also a hugely lower seek latency (orders of magnitute lower; nanoseconds instead of microseconds). If all availble RAM is being used, then swapping will take place, and it will be slow regardless of what filesystem is being used. If all available RAM is not being used, then the filesystem doesn't come into play because swapping isn't taking place at all.10. Swap. Let's just say now, with XFS it actually does feel like I have 4 GB of RAM. Swap slowdown, seems to have dissapeared.
At this point, I think it's fair to acknowledge that for the OP it's a totally different experience. For anyone else not suffering the apparent IO problems that the OP is, the change to XFS will have a very marginal impact, and one which probably won't even be noticed.11. Multimedia editing (something I do everyday with video and audio). Simply put, if you do either, make the switch to XFS. It's a totally different experience.
This is just absurd. If the OP wishes to validate this claim, I welcome him to try. Until then, I would maintain that ZFS is the only "real" next generation filesystem, and that all the others have some catching up to do. I can at least support this claim by pointing to useful, next-generation features that ZFS offers, which no other filesystems do.XFS is the only "real" next generation filesystem and the only one I will use from now on.
Very likely; unfortunately, I would hazard a guess that the OP's isn't. Something is causing that IO bottleneck...Your hardware is fine.
On a personal aside, I recently converted my 160 GB storage drive from ReiserFS to JFS. After doing this, my boot time was reduced from 52 seconds to 30 seconds, because there is no longer a zombie reiserfs.fsck process waiting around for 22 seconds while booting. This does not mean that other people will have the same experience (though I have seen some mention on this thread about slow boot times with Reiser). It also doesn't mean that Reiser is a worse filesystem if a 22 second longer boot process is not critical to you. Many people will swear by Reiser. For anyone reading this thread, I would strongly advise that you research the issues involved with filesystems carefully before deciding to switch (unless you have no data that you care about), and take the OP with a large grain of salt.
Last edited by Bnonn; September 20th, 2006 at 01:37 PM. Reason: clarity
Is there any risk/benefit to using ext3 with xfs for the swap?
I made an xfs partition then selected it as the swap. I don't know maybe it switched it or something. I'll check and write back.
As mentioned before, SWAP is its 'own' unique file system sitting in a partition. You have simply converted an XFS partition to SWAP.