Mustn't be RAID 1 and 0 read-performance be similar?

Mustn't be RAID 1 and 0 read-performance be similar?

[Rustedt, Florian]

Hello List,

If the speed on RAID 0 is based on reading out in parallel, then it must be
the same on RAID 1, mustn't it?

On RAID 1, it is possible, to read two blocks in parallel to speed up, too.

I tried to measure this some weeks ago, but i couldn't get over the
read-performance of a single disk on my raid 1, so that means, that the
software-raid does not use this easy possibility to speed up?

Kind regards, Florian
**********************************************************************************************
IMPORTANT: The contents of this email and any attachments are confidential. They are intended for the 
named recipient(s) only.
If you have received this email in error, please notify the system manager or the sender immediately and do 
not disclose the contents to anyone or make copies thereof.
*** eSafe scanned this email for viruses, vandals, and malicious content. ***
**********************************************************************************************

Re: Mustn't be RAID 1 and 0 read-performance be similar?

[Andy Smith]

[-- Attachment #1: Type: text/plain, Size: 674 bytes --]

Hi Florian,

On Thu, Aug 09, 2007 at 09:15:53AM +0200, Rustedt, Florian wrote:
> On RAID 1, it is possible, to read two blocks in parallel to speed up, too.
> 
> I tried to measure this some weeks ago, but i couldn't get over the
> read-performance of a single disk on my raid 1, so that means, that the
> software-raid does not use this easy possibility to speed up?

Were you doing this with a single thread or multiple threads?  My
understanding is that for RAID-1, a single thread's IO will come
from whichever component is knwon to be closest to where it needs to
start reading from.

If that's not correct I'd love to hear the reality!

Cheers,
Andy

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Mustn't be RAID 1 and 0 read-performance be similar?

[Bill Davidsen]

Rustedt, Florian wrote:
> Hello List,
>
> If the speed on RAID 0 is based on reading out in parallel, then it must be
> the same on RAID 1, mustn't it?
>
> On RAID 1, it is possible, to read two blocks in parallel to speed up, too.
>
> I tried to measure this some weeks ago, but i couldn't get over the
> read-performance of a single disk on my raid 1, so that means, that the
> software-raid does not use this easy possibility to speed up?
>   

We've had this discussion before, and the RAID code does not take 
advantage of running reads in parallel to satisfy readahead or anything 
else. I seen it said that this is on a per-thread basis, but a simple 
test suggests that it's on a process basis, setting two threads to read 
alternating 10k blocks as fast as possible ran no faster than a single 
thread. However, running against RAID-10, even a single thread seemed to 
run far faster than the speed of a single drive.

Lightly tested, don't take this as the last word.

-- 
bill davidsen <davidsen@tmr.com>
  CTO TMR Associates, Inc
  Doing interesting things with small computers since 1979

Re: Mustn't be RAID 1 and 0 read-performance be similar?

[Mario 'BitKoenig' Holbe]

Rustedt, Florian <florian.rustedt@smartnet.de> wrote:
> If the speed on RAID 0 is based on reading out in parallel, then it must be
> the same on RAID 1, mustn't it?
> On RAID 1, it is possible, to read two blocks in parallel to speed up, too.

It's not that simple.
On RAID0 you can read one single stream of data from all of the disks in
parallel: you read one stream from each disk with each stream containing
completely different data and merge them together to get the original
stream. On RAID1 you can only read exactly the same stream from all of
the disks. Thus, RAID1 cannot provide RAID0's speed-up for a single
stream.
However, if you read multiple streams parallel, RAID1 can do better than
RAID0 because you can read stream1 from disk1, stream2 from disk2 etc.
Using RAID0, this speed-up can only be achieved for streams <= chunk
size.


regards
   Mario
-- 
Tower: "Say fuelstate." Pilot: "Fuelstate."
Tower: "Say again." Pilot: "Again."
Tower: "Arghl, give me your fuel!" Pilot: "Sorry, need it by myself..."