xfs
[Top] [All Lists]

Re: nfs/local performance with software raid5 and xfs and SMP

To: "linux-xfs@xxxxxxxxxxx" <linux-xfs@xxxxxxxxxxx>
Subject: Re: nfs/local performance with software raid5 and xfs and SMP
From: Jani Jaakkola <jjaakkol@xxxxxxxxxxxxxx>
Date: Mon, 23 Jul 2001 17:34:14 +0300 (EEST)
In-reply-to: <Pine.LNX.4.33.0107231318310.4170-100000@xxxxxxxxxxxxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
On Mon, 23 Jul 2001, Jani Jaakkola wrote:

> Hmm, I'll guess I have to get my hands on some other benchmark software
> than bonnie and try it with ext2, xfs+internal log and xfs+external log.

OK, it seems that create/delete performance on linux XFS with sw RAID is
still bad with current cvs kernel (not terrible anymore, but bad enough
that I would not put xfs with sw RAID to production use). I downloaded
bonnie++ and did some tests

This is an IDE drive with ext2 (without DMA, since the kernel turns it
off with my IDE controller for some readon):

-----snip------
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
other-fs         2G  5970  93  8058   6  2981  12  5774  80  8553  63 109.1   2
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16   912 100 +++++ +++ +++++ +++   925  99 +++++ +++  5245  99
-----snip------

This is the same IDE drive with xfs (without DMA also!). Not too bad when
compared to ext2:

-----snip------
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
other-fs         2G  7691  99 11853  11  3999  31  6317  85  8652  65 112.4   3
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16   730  57 +++++ +++   781  61   735  54 +++++ +++   366  26
-----snip------

And this is the RAID array with internal log

-----snip------
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
other-fs         2G 11253  96 39684  34  7829   8 10858  89 131226  83 320.2   2
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16    75   1 +++++ +++   114   2    91   2 +++++ +++    61   1
-----snip------

So the RAID array is 5 times faster in pure write IO, but xfs with
internal log kills create/delete performance with IO stalls. With external
log the create/delete performance is again acceptable:

-----snip------
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
other-fs         2G 11235  96 43660  36  7728   8 10891  90 132525  81 322.1   2
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16   854  66 +++++ +++   793  66   833  64 +++++ +++   691  58
-----snip------

So the sw RAID internal log performance is not fixed yet :(

- Jani


<Prev in Thread] Current Thread [Next in Thread>