Re: testing raid 0

[Greg Oster <oster@cs.usask.ca>]

On misc@openbsd.org "Peter Galbavy" <peter.galbavy@knowtion.net> said:
> 
> Maybe you have misunderstood the benefit of RAID 0/1/5 etc. RAID 0
> givesd you the ability to concatenate disks to get a bigger partition
> than would be possible using a single physical device. It will give you
> no performance benefits.

Ummmmmm... "Huh"??  RAID 0 is simple striping across all disks.  RAID 0 in 
RAIDframe does *not* do concatenation.   Even with abysmal controllers+disks 
you should normally see a performance improvement with RAID 0.  With good 
disks, controllers, etc. you should typically see close to a linear speedup.  

> In fact RAID 0 in RAIDframe is probably rather
> inefficient, and you should consider using the simpler ccd(4) drivers.

You have benchmarks to back up this statement?  How about:

              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
RAID 0   1024 11991 92.9 16924 36.9  3110  8.5 10025 92.0 18904 53.5  33.3  2.1
CCD      1024 13035 96.2 17690 34.1  3130  6.7 10594 92.8 18880 43.9  33.5  1.3

Four 2GB FW SCSI drives, split equally over 2 controllers.  Stripe widths on 
ccd and RAID0 were the same.  "looks pretty efficient to me", especially given 
all the other stuff it's doing...

*grumble* 

Later...

Greg Oster