Christian Smith writes:
> Hi,
> I'm having a play with RAIDframe, as I've recently acquired a spare 911 
> enclosure with a couple of seagate barracuda disks in there.
> 
> My plan was to try RAID 0 configuration to see what performance I can get.
> 
> Unfortunately, the aggregate performance using RAID 0 was ~10 slower than 
> I can manage to get out of both disks without raid.
> 
> My primitive performance testing was was done using dd to read from the
> device to /dev/zero, with block size of 64k, and monitor the performance
> using systat vm 1, to give approximate disk throughput along with CPU
> utilisation.
> 
> With RAIDframe, I could manage >5MB/s from any one disk, or >7MB/s
> throughput using both disks together.
> 
> Finally, I have a seagate cheetah disk in there as well, on the same 
> channel, which could move >7MB/s on it's own.
> 
> >From this it's obvious that the SCSI bus gets saturated at ~7-8MB/s, which
> is reasonable as the controller is a 10MB/s fast SCSI-2 (a combined
> SCSI/ethernet SBUS card.)
> 
> Now, with RAIDframe compiled in,

"compiled in"??  RAIDframe shouldn't affect performance like this just by 
being in the kernel... 

> the cheetah tops out at ~2.2MB/s, and the
> RAIDed barracudas can manage <700KB/s between them! I've tried 1, 16, 32 
> and 64 sector interleaves, but performance is just bad.

You are doing RAID 0 over all 3 disks?  And all 3 are on the same channel, 
right?  

Assuming your drives are sd0, sd1, and sd2, what happens when you do:

 csh
 foreach i (0 1 2)
 dd if=/dev/rsd{$i}c of=/dev/null bs=1m count=1000 &
 end

You'll probably want a stripe width of at least 64. 

> I'm running on an IPX with 52MB RAM. Could it be that the IPX is just not 
> upto the task (sys usage is very high during dd.) I'd have thought that 
> RAID-0 would not be very CPU taxing, certainly no more than straight disk 
> reading. And it doesn't explain the drop in performance of a non-RAIDed 
> disk.

You mean the Cheetah isn't in a RAID set, and when it's the only disk being 
accessed, it's slower than it was before you had RAID in the kernel??? 
 
> Any ideas?
> 
> Christian
> 
> PS. As I've been writing this, I've just noticed that doing dd on a disk 
> pushes lev3 interrupts through the roof! Before RAIDframe, I'd get ~200/s 
> when doing dd, now I get more 1500/s! Would certainly explain the lack of 
> performance. What gives?

My guess is a very small stripe width... Try 64 or even 128.

Later...

Greg Oster