View previous topic :: View next topic |
Author |
Message |
xavier10 Guru
Joined: 19 Jan 2004 Posts: 485 Location: Paris, France
|
Posted: Tue Oct 21, 2008 9:03 am Post subject: SSD: which filesystem ? |
|
|
Hello,
I am considering an SSD based laptop, which I would install Gentoo on. I am wondering about support issues with SSD drives and Linux, and what decisions should be made.
In particular, what about the filesystems ? Are all filesystems ok for SSDs or are some preferable to others ?
How do they impact on wear of the SSD ?
Another thing I feel bad about is the swap: is it ok to swap on an SSD ? It seems to me this is a very bad decision due to the wear it may incur on the drive ?
For that reason, I am considering having no swap at all (and taking a decent amount of RAM to begin with, of course!). |
|
Back to top |
|
|
massimo Veteran
Joined: 22 Jun 2003 Posts: 1226
|
|
Back to top |
|
|
jesnow l33t
Joined: 26 Apr 2006 Posts: 895
|
Posted: Wed Oct 22, 2008 9:23 am Post subject: Re: SSD: which filesystem ? |
|
|
The idea of the SSD is that it mimics, and can mimic a normal drive. I used ext3 and it worked fine, I get 123MB/sec throughput on my OCX ssd, as opposed to 80MB/sec on my samsung spinpoint 500GB sata drive. Some things are *much* faster though, and watching a silent emerge --sync is cool.
BUT I'd be curious too what the *optimum* fs would be.
xavier10 wrote: | Hello,
I am considering an SSD based laptop, which I would install Gentoo on. I am wondering about support issues with SSD drives and Linux, and what decisions should be made.
In particular, what about the filesystems ? Are all filesystems ok for SSDs or are some preferable to others ?
How do they impact on wear of the SSD ?
Another thing I feel bad about is the swap: is it ok to swap on an SSD ? It seems to me this is a very bad decision due to the wear it may incur on the drive ?
For that reason, I am considering having no swap at all (and taking a decent amount of RAM to begin with, of course!). |
|
|
Back to top |
|
|
drescherjm Advocate
Joined: 05 Jun 2004 Posts: 2790 Location: Pittsburgh, PA, USA
|
Posted: Wed Oct 22, 2008 3:44 pm Post subject: |
|
|
Quote: | Another thing I feel bad about is the swap: is it ok to swap on an SSD ? It seems to me this is a very bad decision due to the wear it may incur on the drive ? |
I would not worry about this. All descent SSDs have wear leveling. And this will generally increase the life of the device to 3 years or more of continuous (24/7) writing to the drive. _________________ John
My gentoo overlay
Instructons for overlay |
|
Back to top |
|
|
Dagger Retired Dev
Joined: 11 Jun 2003 Posts: 765 Location: UK
|
Posted: Thu Oct 23, 2008 10:34 am Post subject: |
|
|
I'm using 4xSSD (OCZ 128GB) in raid (1+0) array for postgresql database.
SSD aren't super fast in constant reads (only around ~250MB/sec) comparing to SAS array (~320MB/sec), but random access time is awesome (~2ns comparing to ~12ns)
I've got over 500m records in my database and very complicated queries take around ~10 min on SAS array and ~4 min on SSD.
I've been testing these drivers before I put them into the array, and they behave very similar to traditional drives. So generally speaking FS choice should depend on your needs.
I needed very fast random access, so I've chosen reiserfs. For general use ext or xfs should be good.
As drescherjm mentioned I wouldn't really care about lifetime because current generation of drives have longer lifetime than average laptop. _________________ 95% of all computer errors occur between chair and keyboard (TM)
Join the FSF as an Associate Member!
Post under CC license. |
|
Back to top |
|
|
pdw_hu Apprentice
Joined: 02 Jun 2008 Posts: 200 Location: Budapest, Hungary
|
Posted: Thu Mar 05, 2009 5:25 pm Post subject: |
|
|
I'll be getting an OCZ SSD for my laptop soon.
1. What do you think about Ext4 w/o journal? As it's a laptop the only data loss it might endure is near lock-ups, but sysrq can usually at least do a sync before rebooting it.
2. Move /usr/portage to the now freed HDD connected through USB (or FireWire).
3. Mount /var/tmp/portage into tmpfs.
4. Using noatime.
Any other tips? |
|
Back to top |
|
|
OneOfOne Guru
Joined: 28 May 2003 Posts: 368
|
Posted: Thu Mar 05, 2009 6:36 pm Post subject: |
|
|
I'd recommend btrfs and mount with compress,ssd. |
|
Back to top |
|
|
s4e8 Guru
Joined: 29 Jul 2006 Posts: 311
|
Posted: Fri Mar 06, 2009 8:31 am Post subject: |
|
|
OneOfOne wrote: | I'd recommend btrfs and mount with compress,ssd. |
btrfs will eat your SSD very quickly. It write data constantly even in an idle system. |
|
Back to top |
|
|
HeissFuss Guru
Joined: 11 Jan 2005 Posts: 414
|
Posted: Tue Mar 24, 2009 12:43 am Post subject: |
|
|
btrfs will also eat your children. It's not very stable atm. |
|
Back to top |
|
|
poly_poly-man Advocate
Joined: 06 Dec 2006 Posts: 2477 Location: RIT, NY, US
|
Posted: Tue Mar 24, 2009 12:46 am Post subject: |
|
|
ext2 or minix. _________________ iVBORw0KGgoAAAANSUhEUgAAA
avatar: new version of logo - see topic 838248. Potentially still a WiP. |
|
Back to top |
|
|
d2_racing Bodhisattva
Joined: 25 Apr 2005 Posts: 13047 Location: Ste-Foy,Canada
|
Posted: Tue Mar 24, 2009 12:53 am Post subject: |
|
|
Ext2, in fact, you don't need to log the change. |
|
Back to top |
|
|
ial Apprentice
Joined: 27 Dec 2008 Posts: 161 Location: Warsaw (Warszawa)
|
Posted: Sat Jul 18, 2009 12:44 am Post subject: |
|
|
and what about Reiser4?
especially with its compression set on -- will the transparent compression move the burden away slightly from stressing the actual physical medium rather into operations located more within buffer/RAM ? I mean, files before a write are compressed significantly in fs buffer so afterall much less resulting data is physically engraved on SSD medium, and much fewer NAND cells are being worn, is that correct?
s4e8 wrote: | btrfs will eat your SSD very quickly. It write data constantly even in an idle system. |
I hope Reiser4 does not behave that ugly...?
BTW. What does the option "SSD Mode" mean in Btrfs and how does it improve SSD wear friendliness? Will Btrfs under this option more or less wear physical media than ext2? |
|
Back to top |
|
|
Need4Speed Guru
Joined: 06 Jun 2004 Posts: 497
|
Posted: Sat Jul 18, 2009 3:27 am Post subject: |
|
|
s4e8 wrote: | OneOfOne wrote: | I'd recommend btrfs and mount with compress,ssd. |
btrfs will eat your SSD very quickly. It write data constantly even in an idle system. |
This is just not true. Modern SSDs have WEAR LEVELING . This means you can write to them constantly 24/7 for many YEARS before they will fail. Most SSDs now have longer MTBFs than traditional hard disks. _________________ 2.6.34-rc3 on x86_64 w/ paludis
WM: ratpoison
Term: urxvt, zsh
Browser: uzbl
Email: mutt, offlineimap
IRC: weechat
News: newsbeuter
PDF: apvlv |
|
Back to top |
|
|
ial Apprentice
Joined: 27 Dec 2008 Posts: 161 Location: Warsaw (Warszawa)
|
|
Back to top |
|
|
Need4Speed Guru
Joined: 06 Jun 2004 Posts: 497
|
Posted: Wed Jul 22, 2009 4:18 am Post subject: |
|
|
I just switched over to NILFS2 for my rootfs and fixed the partition alignment to 512k. It has a made HUGE difference in write speeds!
Random writes used to be glacial and would sometimes cause my system to hang for 10-30 seconds if a fsync were forced.
I am using a "Gen 1" Ridata SSD, so NILFS2 is probably helping me more than if you owned something like Intel's X25-m, which has "better" firmware that tries to hide the some of the SSD's characteristics from the filesystem. But I would still give NILFS2 a try if you want faster writes.
SSD Alignment Info:
http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an-ssds-erase-block-size/
http://www.ocztechnologyforum.com/forum/showpost.php?p=373226&postcount=98 _________________ 2.6.34-rc3 on x86_64 w/ paludis
WM: ratpoison
Term: urxvt, zsh
Browser: uzbl
Email: mutt, offlineimap
IRC: weechat
News: newsbeuter
PDF: apvlv |
|
Back to top |
|
|
energyman76b Advocate
Joined: 26 Mar 2003 Posts: 2048 Location: Germany
|
Posted: Wed Jul 22, 2009 10:22 am Post subject: |
|
|
reiser4 is always the right answer - because of moving journal, compression, it is fast... _________________ Study finds stunning lack of racial, gender, and economic diversity among middle-class white males
I identify as a dirty penismensch. |
|
Back to top |
|
|
Grubshka n00b
Joined: 07 Nov 2004 Posts: 37 Location: Toulouse - France
|
Posted: Wed Aug 12, 2009 9:24 pm Post subject: |
|
|
I read that NILFS was making a lot of disk operations, which was bad for SSD longevity? Is this true?
(This may depend of people means with "longevity" : I don't expect to use my laptop more than 10 years).
I go every days on the webstore, I think I'll click on "buy one day, when I'll decide which one to buy... |
|
Back to top |
|
|
energyman76b Advocate
Joined: 26 Mar 2003 Posts: 2048 Location: Germany
|
|
Back to top |
|
|
wildhorse Apprentice
Joined: 16 Mar 2006 Posts: 150 Location: Estados Unidos De América
|
Posted: Thu Aug 13, 2009 5:29 am Post subject: |
|
|
Patriot Memory is backing its new Torqx M28 Series SSDs with a 10-year warranty. Is there any IDE (PATA/SATA) HDD available with anything close to a 10-year warranty?
About the pagefile, I say no pagefile is the best pagefile. Go for RAM. |
|
Back to top |
|
|
energyman76b Advocate
Joined: 26 Mar 2003 Posts: 2048 Location: Germany
|
Posted: Thu Aug 13, 2009 11:07 am Post subject: |
|
|
you need a 'pagefile' to be able to overcommit. Which you need if you really want to make use of your ram. It doesn't have to be big. Just be there. _________________ Study finds stunning lack of racial, gender, and economic diversity among middle-class white males
I identify as a dirty penismensch. |
|
Back to top |
|
|
bassai n00b
Joined: 13 Jan 2008 Posts: 9 Location: Munich, Bavaria
|
Posted: Thu Aug 13, 2009 2:37 pm Post subject: |
|
|
I'm running ext4 on a SSD. This works fine for 3 months now.
I mounted /var/tmp/portage on tmpfs to compile in RAM.
Furthermore I increased my sync interval. |
|
Back to top |
|
|
pdw_hu Apprentice
Joined: 02 Jun 2008 Posts: 200 Location: Budapest, Hungary
|
Posted: Thu Aug 13, 2009 4:51 pm Post subject: |
|
|
bassai wrote: | I'm running ext4 on a SSD. This works fine for 3 months now.
I mounted /var/tmp/portage on tmpfs to compile in RAM.
Furthermore I increased my sync interval. |
Same here, and am using data=writeback. For some reason s2ram doesn't work (journal problems), but as i don't really need it i haven't dug myself into that issue. |
|
Back to top |
|
|
runem n00b
Joined: 08 Sep 2006 Posts: 40
|
Posted: Thu May 13, 2010 10:06 pm Post subject: |
|
|
pdw_hu wrote: | bassai wrote: | I'm running ext4 on a SSD. This works fine for 3 months now.
I mounted /var/tmp/portage on tmpfs to compile in RAM.
Furthermore I increased my sync interval. |
Same here, and am using data=writeback. For some reason s2ram doesn't work (journal problems), but as i don't really need it i haven't dug myself into that issue. |
Acording to http://patchwork.ozlabs.org/patch/49687/ data=writeback and TRIM do not mix well, so I have changed to data=ordered. Using deadline io-sched wtth fifo_batch=1 seems to work well. |
|
Back to top |
|
|
tnt Veteran
Joined: 27 Feb 2004 Posts: 1227
|
Posted: Thu May 13, 2010 10:39 pm Post subject: |
|
|
runem wrote: | Using deadline io-sched wtth fifo_batch=1 seems to work well. |
have you noticed any performance difference over default fifo_batch=16 ?
I've read somewhere that recent kernels have some ssd-related optimizations for cfs, but unfortunately I've forgot where.
can anyone confirm that? _________________ gentoo user |
|
Back to top |
|
|
runem n00b
Joined: 08 Sep 2006 Posts: 40
|
Posted: Fri May 14, 2010 9:06 am Post subject: |
|
|
tnt wrote: |
have you noticed any performance difference over default fifo_batch=16 ?
I've read somewhere that recent kernels have some ssd-related optimizations for cfs, but unfortunately I've forgot where.
can anyone confirm that? |
I have just made at test with tiobench and the default of 16 was slightly better
If /sys/block/<device>/queue/rotational is set to 0 then the scheduler works better with SSDs and USB-sticks for that matter. |
|
Back to top |
|
|
|