>From the Software-RAID-2.html....
" ...
RAID-0 is much like RAID-linear, except that the component partitions
are divided into stripes and then interleaved. Like RAID-linear, the
result is a single larger virtual partition.
RAID-1 is also referred to as "mirroring". Two (or more) partitions, all
of the same size, each store an exact copy of all data, disk-block by
disk-block.
..."
-----Original Message-----
From: David Harris [mailto:[EMAIL PROTECTED]]
Sent: Saturday, December 26, 1998 9:53 AM
To: Eyal Lebedinsky; [EMAIL PROTECTED]
Subject: RE: software raid chunk size
Hi,
Eyal Lebedinsky wrote:
> I am now studying the RAID implementation. For starters I attached
some
> discarded sun disks (6x200MB QUANTUM Model: PD210S SUN0207 Rev: 492S)
> and set to measure the results. the disks seem to be slow on their own
> (hdparm says 1.36 MB/sec) and I expected some speedup when striped
> (raid0). I got 2.8 MB/sec, just over twice the speed. What gives? I use an
> ASUS SC200 SCSI controller, Narrow SCSI, on 2.1.132 with the latest
> raid patch and tools.
You mention that you are trying to achieve a read speed boost from
"striping", but then mentioned that you are running raid0. Your problem
here
might be that raid0 is not striping, but rather mirroring. It is true
that
mkraid requires a chunk-size for raid0, but this is not used for any
striping. That chunk-size is there just to pacify the internal mechanics
of
the driver. I recommend specifying 128k.
For a description of the different raid levels, see:
http://linas.org/linux/Software-RAID/Software-RAID-2.html
But ignore that raid-HOWTO when it talks about the specifics of the
drivers,
because it was written for a completely different earlier version.
- David Harris
Principal Engineer, DRH Internet Services
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]] On Behalf Of Eyal Lebedinsky
Sent: Saturday, December 26, 1998 8:22 AM
To: [EMAIL PROTECTED]
Subject: software raid chunk size
I am trying to understand the use of this parameter. It says that this
is
the size of a continuous chunk of data on one disk. So I guess one would
want to make it small, spreading data accross disks to gain performance
from the striping.
But this does not seem to be the case. A minimal reasonable size of 8k
means than most one block (1k or 4k) requests will come from one disk
with no striping. How dop I get performance when I do randon
(non-sequential)
reads from a file? If i read each block as a number of small (say 512b
sectors)
reads in parallel from a number of disks then I see how I gain.
I must be missing somethging simple here. I did read the doco which not
enough. I checked the list archives but could not do a search. I read
the
last few collections and could not find any related item.
I am now studying the RAID implementation. For starters I attached some
discarded sun disks (6x200MB QUANTUM Model: PD210S SUN0207 Rev: 492S)
and set to measure the results. the disks seem to be slow on their own
(hdparm says 1.36 MB/sec) and I expected some speedup when striped
(raid0). I
got 2.8 MB/sec, just over twice the speed. What gives? I use an ASUS
SC200
SCSI controller, Narrow SCSI, on 2.1.132 with the latest raid patch and
tools.
I played with the chunk size (from a low of 4k to a high of 32k) with no
visible improvement.
--
Eyal Lebedinsky ([EMAIL PROTECTED])
