Re: journaled file systems
On Sat, Aug 11, 2001 at 11:59:30AM +0800, Donald Szeto wrote:
> So there remains only ReiserFS and XFS for choice.
>
> ReiserFS isn't bad, but XFS presently shows faster performance than
> ReiserFS.
actually, it doesn't.
i've done extensive benchmarking of xfs and reiserfs over the last month
or so and my results show that reiserfs is much faster than xfs for
nearly everything but large block reads - and even there it isn't that
far behind.
for the lots-of-little-files usage pattern associated with mail spools,
mail queues, Maildir/ directories, news spools, etc, reiserfs blows XFS
away.
i use both reiserfs and XFS. for mail (and news) servers, i use
reiserfs. for everything else, i use xfs.
here's what bonnie has to say about xfs and reiserfs (with various mkfs
and mount time options):
Version 1.01b ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
xfs 1G 13019 99 65285 34 16726 12 9041 66 33448 9 562.1 1
xfs logbufs=8 1G 13031 99 66354 34 16504 12 9008 66 33645 10 562.0 2
xfs lb=8 osyncis 1G 13122 99 66346 36 16292 11 8981 66 33306 10 564.5 2
xfs extlog 1G 13048 99 72007 41 16170 13 9064 67 33247 11 575.1 2
xfs sunit=256 sw 1G 12701 97 68047 39 16446 12 10043 80 48239 15 566.2 2
xfs final 1G 13002 99 74797 43 15664 12 10078 80 47764 17 528.3 2
reiserfs 1G 12287 99 63942 63 18234 14 10058 75 44877 15 559.4 2
reiserfs notail 1G 12305 99 63196 62 17312 13 10432 78 42547 14 554.5 2
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
xfs 16 4075 58 +++++ +++ 3608 50 4017 63 +++++ +++ 2815 45
xfs logbufs=8 16 4601 82 +++++ +++ 4000 65 4495 82 +++++ +++ 3051 56
xfs lb=8 osyncis 16 4292 79 +++++ +++ 4591 68 3357 59 +++++ +++ 3353 58
xfs extlog 16 463 26 +++++ +++ 432 21 462 27 +++++ +++ 390 20
xfs sunit=256 sw 16 4363 90 +++++ +++ 3860 69 4333 91 +++++ +++ 2786 55
xfs final 16 3927 69 +++++ +++ 3427 49 3440 58 +++++ +++ 2427 42
reiserfs 16 14588 99 +++++ +++ 16403 100 13459 98 +++++ +++ 13398 99
reiserfs notail 16 14447 99 +++++ +++ 15461 94 13823 100 +++++ +++ 13520 100
these tests were all run on the same hardware (p3-933 with 512MB ram,
hardware raid5 with 128MB non-volatile cache ram. drives were 8 x 72GB
U160 cheetah scsi drives.)
notes:
extlog == external log device (in this case, an ATA100 drive).
performance really sucked.
osyncisdsync is as per the xfs mount option described in mount(1). makes
little difference to performance, may be safer for mail serves.
"xfs sunit=256" is the result of telling mkfs.xfs the sunit and swidth
of my raid5 device. these tweaks make an enormous difference for XFS.
"xfs final" is basically the same as "xfs sunit=256" but after the
server had been rebuilt in it's final production configuration, with
other stuff running (not doing much). also with an smp kernel, so the
2nd cpu accounts for the difference in %CPU usage. this machine is going
to be the file server for a web farm, so XFS is the right fs for it.
> Furthermore, XFS has many features such as ACL, etc.
yes. XFS has great ACLs, quota support, and works out of the box with
NFS.
it's also had a few years more real-world testing (on SGI's IRIX, not
linux) than reiserfs.
overall, i think i prefer xfs. but reiserfs still has it's place.
> For e-mail servers, there will be numerous file activities.
> XFS should be your better choice than ext3.
yep. and, IMO & IME, reiserfs is an even better choice for mail servers.
craig
--
craig sanders <cas@taz.net.au>
Fabricati Diem, PVNC.
-- motto of the Ankh-Morpork City Watch
Reply to: