| View previous topic :: View next topic |
| Author |
Message |
oegat
n00b
Joined: 12 Apr 2003
Posts: 41
Location: Sweden
|
 Posted: Fri Jan 11, 2008 3:22 pm Post subject: XFS on amd64? |
 |
|
The installation handbook for amd64 gentoo does not recommend using XFS, it claims that it is not sufficiently tested. When searching the forums, I find people using it but also people having problems: https://forums.gentoo.org/viewtopic-t-617720-highlight-xfs+amd64.html
You who have tried around, do you recommend against using XFS on an amd64 system (mine is an athlon x2 with ati sb600 chipset) or would you say it seems safe (from your experience)?
I'd use it for a storage of media files, nothing really critical, but anyway.
_________________
Högt över fjällen där flyger en ko
kanske den flyger på sin tro? |
|
| Back to top |
|
 |
Spooky Ghost
Apprentice
Joined: 19 Apr 2002
Posts: 210
Location: Bristol, United Kingdom
|
| Posted: Fri Jan 11, 2008 5:12 pm Post subject: |
|
|
| I've been using XFS for several years and rarely encountered problems. My machine is mostly used for serving files but I recently started using XEN and I have a few domU instances using XFS without problems. The only trouble I have experienced is after a crash I've once or twice had to run xfs_repair and there was also a subtle kernel bug which slipped through in recent history but that was quickly caught and only required an xfs_repair to fix. My personal opinion is that it is solid and reliable. YMMV of course:) |
|
| Back to top |
|
|
Urban Cowboy
n00b
Joined: 09 Oct 2007
Posts: 64
|
| Posted: Fri Jan 11, 2008 5:38 pm Post subject: |
|
|
I've used it once and saw no real huge benefit over JFS or reiser. So if there are supposedly issues - I wouldn't bother.
_________________
Anything worth doing is worth over-doing. Moderation is for cowards. |
|
| Back to top |
|
|
oegat
n00b
Joined: 12 Apr 2003
Posts: 41
Location: Sweden
|
| Posted: Fri Jan 11, 2008 5:55 pm Post subject: |
|
|
Spooky Ghost:
Ok! And you use it under amd64 now? On which hardware and for how long on that hardware?
_________________
Högt över fjällen där flyger en ko
kanske den flyger på sin tro? |
|
| Back to top |
|
|
overkll
Veteran
Joined: 21 Sep 2004
Posts: 1249
Location: Austin, Texas
|
| Posted: Fri Jan 11, 2008 7:13 pm Post subject: |
|
|
If you have no specific reason to use XFS, I'd recommend sticking with ext2/ext3 with the "dir_index" option enabled for a standard gentoo install. It's reliable, fast, and has low overhead. I don't know why the gentoo handbook doesn't mention the option. Fedora uses it by default and I believe some other distros do as well.
There is a massive post regarding ext2/3 somewhere in these forums that goes into more detail about the option as well as some lesser used options for ext2/3. I believe that post also contains a link to an external site which shows comprehesive performance testing results between XFS, JFS, reiserfs, and ext2/3. IIRCC, ext with dir_index was the best in overall general performance. I'll see if I can find it. If I can, I'll post a link.
Personally, this is my opinion of the filesystems I've tried:
For general use, ext2 and 3 with dir_index option.
For giamongous files (> 1 Gb) JFS may be better.
For giamongous partitions (> 1 Tb), XFS may be your best option.
Avoid reiserfs like the Plague!
To format with the "dir_index" option:
ext2:
| Code: | | # mke2fs -O dir_index /dev/<your partition> |
ext3:
| Code: | | # mke2fs -O has_journal,dir_index /dev/<your partition> |
I have a mythtv box, and I use JFS for the media partition, ext2 with dir_index for boot, and ext3 with dir_index for the rest. |
|
| Back to top |
|
|
Habbit
Apprentice
Joined: 01 Sep 2007
Posts: 237
Location: 3.7137 W, 40.3873 N
|
| Posted: Fri Jan 11, 2008 8:37 pm Post subject: |
|
|
| overkll wrote: | | Avoid reiserfs like the Plague! |
Why should one do so? I run 3 Gentoo systems on reiserfs (root, /var, /tmp, etc) and they are all fine. Don't get me wrong, I'm not a reiserfs zealot either (another computer I have runs on Ubuntu ext3), it's just I'd like to know the technical reasons behind such a categoric statement.
_________________
| Code: | ~ $ objdump -d ./habbit_mind
90 xchg %rax, %rax
EB FD jmp $-3 |
|
|
| Back to top |
|
|
overkll
Veteran
Joined: 21 Sep 2004
Posts: 1249
Location: Austin, Texas
|
| Posted: Fri Jan 11, 2008 8:54 pm Post subject: |
|
|
Don't mean to start a flame war, just stating my opinion.
Over time, reiserfs fragments like a sonofabitch. That, in turn, affects the efficiency and speed. If you've had portage on reiserfs, over time, you will see your "emerge --sync" s metadata updates slow down to a crawl. I had two AMD64 machines running reisers - one in 32 bit mode - and after a year or so, the metadata portion of the "emerge -sync" took up to a half-hour! Migrated both to ext3 with dir_index and now the metadata portion only takes about 1 minute. After two years, it's still around 1 minute.
I've also had reiserfs crash unrecoverably, TWICE! Not so with ext3 (not yet anyway).
BTW, the metadata portion of an emerge --sync is the
| Code: | | >>> Updating Portage cache: 31% |
portion at the end of the rsync' ing process.
EDIT: Food for thought...
If reiserfs is so great and stable, why don't the major linux distro's use it by default? |
|
| Back to top |
|
|
Habbit
Apprentice
Joined: 01 Sep 2007
Posts: 237
Location: 3.7137 W, 40.3873 N
|
| Posted: Fri Jan 11, 2008 9:23 pm Post subject: |
|
|
| overkll wrote: | | I've also had reiserfs crash unrecoverably, TWICE! Not so with ext3 (not yet anyway). |
What do you mean by "crash unrecoverably"? Something that even fsck --rebuild-tree could not repair? I have heard of those failures, though I have yet to encounter one. It's true that ext3 is more stable, though. My main use for reiserfs is in laptops and computers with small HDs, where I use it for partitions with lots of small files: /usr/portage, /var/tmp and /tmp are among the first, but also /var/log if you have many log-producing apps. For instance, a full Portage tree (without distfiles) takes nearly 500 MiB in an ext3 drive: with reiserfs (without notail, of course) they can be cramped into ~260MiB.
Back to the main post: As I said, I'd advocate reiser (without the "notail" option) for a partition with large numbers of small files (can save space, won't waste space on unused inodes or run out of them), but for media files (MP3s, xvid movies ~700MiB), just about any filesystem is good. If by "media files" you mean movies and files > 5 GiB, however, then XFS is a good choice, since its unlink() (removal) operation seems to be faster than ext3 (mythtv used to have a "slow deletion" option for ext3 filesystems, so that the pc would not freeze while deleting large files).
As a side advise: if you have a bit of RAM to spare (say, 10 MiB), you can replace /var/run and /var/lock with tmpfs filesystems, which could improve the performance of lock-intensive processes. Also, if your computer crashes, there will be no garbage sitting in those (inherently volatile) directories to confuse your system on next bootup.
_________________
| Code: | ~ $ objdump -d ./habbit_mind
90 xchg %rax, %rax
EB FD jmp $-3 |
|
|
| Back to top |
|
|
BitJam
Advocate
Joined: 12 Aug 2003
Posts: 2508
Location: Silver City, NM
|
| Posted: Fri Jan 11, 2008 9:31 pm Post subject: |
|
|
DRobbins posted an interesting article about file systems. It is mostly about Ext3 but it also mentions that XFS does not journal data so data corruption is possible with XFS.
I use a small-block XFS file system for /usr/portage (minus distfiles) and it has great performance, often syncing in about one minute. But I use Ext3 for most of my other file systems. I also backup regularly. |
|
| Back to top |
|
|
overkll
Veteran
Joined: 21 Sep 2004
Posts: 1249
Location: Austin, Texas
|
|
| Back to top |
|
|
overkll
Veteran
Joined: 21 Sep 2004
Posts: 1249
Location: Austin, Texas
|
| Posted: Fri Jan 11, 2008 9:35 pm Post subject: |
|
|
| Habbit wrote: | | What do you mean by "crash unrecoverably"? Something that even fsck --rebuild-tree could not repair? |
Yes, that's exactly what I mean. And I'm not the only one. |
|
| Back to top |
|
|
Habbit
Apprentice
Joined: 01 Sep 2007
Posts: 237
Location: 3.7137 W, 40.3873 N
|
| Posted: Fri Jan 11, 2008 10:27 pm Post subject: |
|
|
| BitJam wrote: | | it also mentions that XFS does not journal data so data corruption is possible with XFS. |
Nor does ext3 with the default options, it only journals metadata, though you can enable full data journaling with a mount option.
_________________
| Code: | ~ $ objdump -d ./habbit_mind
90 xchg %rax, %rax
EB FD jmp $-3 |
|
|
| Back to top |
|
|
BitJam
Advocate
Joined: 12 Aug 2003
Posts: 2508
Location: Silver City, NM
|
| Posted: Fri Jan 11, 2008 11:54 pm Post subject: |
|
|
| Habbit wrote: | | BitJam wrote: | | it also mentions that XFS does not journal data so data corruption is possible with XFS. |
Nor does ext3 with the default options, it only journals metadata, though you can enable full data journaling with a mount option. |
I agree that you can specify full data journalling as a mount option but the default setting protects your data without the extra overhead of full data journalling.
The DRobbins article I linked to says: | Quote: | | More interestingly, ext3 also offers another journaling mode that provides the benefits of full journaling but without introducing a severe performance penalty. ... (called "data=ordered" mode) |
The Ext Filesystems Tips someone else linked to says: | Quote: | | By default, ext3 partitions mount with the 'ordered' data mode |
Therefore, unless one of those two sources is wrong, ext3 by default protects your data as well as the metadata. If you have evidence to the contrary, please let us know. |
|
| Back to top |
|
|
oegat
n00b
Joined: 12 Apr 2003
Posts: 41
Location: Sweden
|
| Posted: Sat Jan 12, 2008 12:34 am Post subject: |
|
|
Ok, there seems to exist lots of interesting reading regarding the performance of different fs. I've read some of it and I'll read more when I have time.
I'm anyway using LVM2 so I think I will use ext3 primarily to start out with but then allocate a moderate-size (20G or so) xfs in the same volume group to play around with. If I'd find xfs superior I may increase it incrementally later. (since ext3 may be shrunk but not xfs).
Using tmpfs for /var/run and /var/lock sounds like a good idea, I think that I will try it.
And filesystem crashes, well we all have our own stories. My only significant fs crash was actually with ext3, the first time I tried it out ever. Before that I had used ext2 for years without problems. I don't remember the whereabouts of the crash, I think it was my root partition after or during a fresh install - dont remember if I fixed it with fsck or if I simply recreated it and reinstalled. Not that it scared me off from using ext3 later, it remained my preferred choice for personal non-expendible data anyway afterwards (and still is).
_________________
Högt över fjällen där flyger en ko
kanske den flyger på sin tro? |
|
| Back to top |
|
|
Habbit
Apprentice
Joined: 01 Sep 2007
Posts: 237
Location: 3.7137 W, 40.3873 N
|
| Posted: Sat Jan 12, 2008 2:00 am Post subject: |
|
|
| BitJam wrote: | The Ext Filesystems Tips someone else linked to says: | Quote: | | By default, ext3 partitions mount with the 'ordered' data mode |
Therefore, unless one of those two sources is wrong, ext3 by default protects your data as well as the metadata. If you have evidence to the contrary, please let us know. |
data=ordered only journals metadata, but also takes the extra caution of _ordering_ the data writes to happen (and a "flush" command to be sent) before the metadata "transaction" is marked as "done" in the journal. This should indeed be safer than usual, metadata-only journaling, in which out-of-order writes could cause the metadata to be send to the plates before the data is, thus adding garbage to files which were being appended to. However, even the kernel cannot _fully_ control when the drive actually commits its caches to disk, so this level of security is relative. In some drives it will be real, in some others it won't.
_________________
| Code: | ~ $ objdump -d ./habbit_mind
90 xchg %rax, %rax
EB FD jmp $-3 |
|
|
| Back to top |
|
|
aceFruchtsaft
Guru
Joined: 16 May 2004
Posts: 438
Location: Vienna, Austria
|
| Posted: Sat Jan 12, 2008 1:01 pm Post subject: |
|
|
| overkll wrote: |
I've also had reiserfs crash unrecoverably, TWICE! Not so with ext3 (not yet anyway). |
Sorry, but this is completely irrelevant. You will find tons of people who had problems with <insert any filesystem here> on the internet.
I've had one broken ext3 and one broken reiserfs partition, but besides that, I've been using reiserfs on ~ 15 partitions on 3 Gentoo systems for years without problems (also in more complex setups with Device Mapper + LVM2 + reiserfs or hardware RAID + LVM2 + reiserfs).
BTW, the only reason why openSUSE does not use reiserfs as the default FS is due to a lack of maintainers, not because of its technical aspects.
Concerning the OP's question:
I've also been using DM + LVM2 + XFS on amd64 without problems so far, but not for a very long time, so I don't know yet whether to recommend it or not. |
|
| Back to top |
|
|
aceFruchtsaft
Guru
Joined: 16 May 2004
Posts: 438
Location: Vienna, Austria
|
| Posted: Sat Jan 12, 2008 1:06 pm Post subject: |
|
|
| Habbit wrote: | | As a side advise: if you have a bit of RAM to spare (say, 10 MiB), you can replace /var/run and /var/lock with tmpfs filesystems, which could improve the performance of lock-intensive processes. Also, if your computer crashes, there will be no garbage sitting in those (inherently volatile) directories to confuse your system on next bootup. |
BTW, cleaning /var/lock and /var/run at startup is taken care of by the init scripts on any reasonable linux system. Have a look at /etc/init.d/bootmisc. |
|
| Back to top |
|
|
nbkolchin
Apprentice
Joined: 07 Feb 2004
Posts: 290
Location: Russia, Saint-Petersburg
|
| Posted: Sat Jan 12, 2008 2:55 pm Post subject: Re: XFS on amd64? |
|
|
| oegat wrote: | | You who have tried around, do you recommend against using XFS on an amd64 system (mine is an athlon x2 with ati sb600 chipset) or would you say it seems safe (from your experience)? |
I'm using XFS as a primary partition for two years on amd64 system. No problems.
Once one of harddisks got a bad sector in the xfs journal area and fs refuses to mount. But xfstools helped me to recover all data with little problems (I didn't know that XFS created on 64-bit arch, can't be mounted on 32-bit).
P.S. http://www.everything2.com/index.pl?node_id=1479435 <-- I used this guide to tune XFS for everyday work. |
|
| Back to top |
|
|
DragonK
n00b
Joined: 24 Feb 2005
Posts: 40
|
| Posted: Fri Jan 18, 2008 7:16 am Post subject: |
|
|
I've also been using XFS on amd64 for more than 8 months now and no problems so far. 
However, I've also tried XFS on a dmraid setup (amd64, raid 1, intel storage matrix...) and it started to report strange errors when checked...so I wouldn't recommend it in such cases (can't exclude hardware problems though....yet.).. |
|
| Back to top |
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
|
Other Things Gentoo
