Re: Tweaking/Optimizing MD RAID: 195MB/s write, 181MB/s read (so far)

[Justin Piszcz <jpiszcz@xxxxxxxxxxxxxxx>]

RAID5 with 128kb chunk size.

Raid0 was 317MB/s write and 279MB/s read.

# xfs_growfs -n /dev/md3
meta-data=/dev/md3               isize=256    agcount=16, agsize=6868160 
blks
         =                       sectsz=4096  attr=0
data     =                       bsize=4096   blocks=109890528, imaxpct=25
         =                       sunit=32     swidth=96 blks, unwritten=1
naming   =version 2              bsize=4096  
log      =internal               bsize=4096   blocks=32768, version=2
         =                       sectsz=4096  sunit=1 blks
realtime =none                   extsz=393216 blocks=0, rtextents=0


On Fri, 12 Jan 2007, David Chinner wrote:

> On Thu, Jan 11, 2007 at 06:05:36PM -0500, Justin Piszcz wrote:
> > With 4 Raptor 150s & XFS (default XFS options):
> 
> I need more context for this to be meaningful in any way.
> 
> What type of md config are you using here?  RAID0, 1, 5.....?
> 
> What's the raw device throughput (i.e. read and write to /dev/md3)?
> 
> output of 'growfs_xfs -n <mntpt>'? (i.e. is XFS doing aligned
> or unaligned allocation/IO)?
> 
> If it's raid0, how does it compare with a dm stripe?
> 
> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> Principal Engineer
> SGI Australian Software Group
>