Re: Linux Software RAID5 - ballpark numbers
- From: anton@xxxxxxxxxxxxxxxxxxxxxxxxxx (Anton Ertl)
- Date: Wed, 31 Jan 2007 16:20:56 GMT
Rohan Beckles <rohan.beckles@xxxxxxxxxx> writes:
Firstly, I done some individual tests on the drives, measuring the MB/s
unbuffered (sync option on the mount command) write speed.
Is the sync mount option the normal way you will be using the file
system? If not, better don't use that mount option, and measure write
performance by timing the write followed by a sync, e.g.:
time sh -c 'dd if=/dev/zero of=bla bs=64k count=16384; sync'
My guess is that you see bad interaction between whatever a
sync-mounted file system does and the md driver (probably lots of
partial chunk writes etc.).
Now, I realise that having different disks from manufacturers doesn't help,
as there are clear performance discrepancies between the individual units,
but does that really explain such low performance?
We require that every disk is from a different manufacturer, to reduce
the risk of systematic multiple-disk failures, but we use only two
disks per RAID. I just saw 89MB/s write performance on a two-disk
RAID0 through ext3 and 47MB/s on a two-disk RAID1 with the test above.
The disks were 400GB SATA drives from Seagate and WD.
- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@xxxxxxxxxxxxxxxxxxxxxxxxxx Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html
.
- Follow-Ups:
- Re: Linux Software RAID5 - ballpark numbers
- From: Rohan Beckles
- Re: Linux Software RAID5 - ballpark numbers
- References:
- Linux Software RAID5 - ballpark numbers
- From: Rohan Beckles
- Linux Software RAID5 - ballpark numbers
- Prev by Date: Re: Linux Compatible mp3 players other than ipods?
- Next by Date: Re: Linux Software RAID5 - ballpark numbers
- Previous by thread: Re: Linux Software RAID5 - ballpark numbers
- Next by thread: Re: Linux Software RAID5 - ballpark numbers
- Index(es):
Relevant Pages
|