lkml.org 
[lkml]   [1999]   [Jan]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: RAID0: performance oddity (2.2.0pre7)
On Sun, 17 Jan 1999, Chris Evans wrote:

> I have a raid0 array of 2 disks, both of which can do ~6.5Mb/sec.
>
> With chunksize = 4k (default), the md0 device can shift ~11.5Mb/sec. Not
> bad.
>
> However bumping up the chunksize to 64k _decreases_ performance on the
> array to ~6Mb/sec. This is slower than either individual disk!
>
> I would have thought increasing chunksize would increase performance due
> to decreased overhead?

It all depends on the caching and read-ahead strategies used by
the drives. Some drives are happy if you read every other 4k,
some are happy if you read every other 32k, some are happy if
you read every 128k... (IBM's seem to be generally happy, Seagate's
don't even raise a smile unless you use 32k). As soon as the drive
takes a read miss on its cache you have to wait for the physical
I/O to happen - which increases the access time dramatically.

Warning 1: When setting up a RAID array you *need* to experiment
with chunk sizes otherwise you may well be wasting your time in
a big way.

Warning 2: If you mix different drives in an array there may be
*no* chunk size which works well for all drives and it may be
impossible to get the array to work even as well as a single drive.

Warning 3: If you use the same drives they probably have the
*same* failure modes. If you bought them together they likely
come from the *same* batch as well. Don't be surprised if they
all go bad at much the same time :-).

Warning 4: RAID is not a panacea :-).

Mike

--
A train stops at a train station, a bus stops at a bus station.
On my desk I have a work station...
.----------------------------------------------------------------------.
| Mike Jagdis | Internet: mailto:mike@roan.co.uk |
| Roan Technology Ltd. | |
| 54A Peach Street, Wokingham | Telephone: +44 118 989 0403 |
| RG40 1XG, ENGLAND | Fax: +44 118 989 1195 |
`----------------------------------------------------------------------'


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:49    [W:0.045 / U:0.212 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site