complain about dpt_2o driver and CPU time

complain about dpt_2o driver and CPU time

[David Chow]

Dear Deanna and others,

I am using the ASR2100 RAID card and running 2.4.17 Linux . It seems the 
card is very slow in RAID5 computation. I use 5x18GB 10,000rpm seagate 
drives. The card orginal comes with a 32MB ECC SD RAM as cache buffer. I 
feel that the card is very slow, the test with Bonnie shows 1810KB/sec 
with CPU load less than 2%. This shows the CPU time is idle, but the 
load average go up to around 13-14 . This means the dpt_i2o driver is 
probably not written very good in releasing the cpu time (call to 
schedule()) on blocking calls, probably in some IO waiting routines(I am 
not sure), or it is holding some big locks in a slow place. The CPU is 
sitting there doing nothing, but this behaviour always block my NFS 
clients ("server not responding, still trying") and the machine looks 
like dead not even response to a console keyboard input. Can you tell me 
about what is going on or who else is suffering from the same result? I 
would like to do some improvement of the driver, or the card is really 
that slow? My 10year old SCSI drive (no raid) runs 2200Kb/sec and is 
even faster than this Ultra-160 SCSI i960 RAID with 32MB cache. If this 
is the case, I will just throw it away and use my onboard Ultra-160 
connector with software RAID which runs faster(at least). This is almost 
unusable for even small workgroup server.

regards,
David

Re: complain about dpt_2o driver and CPU time

[Alan Cox]

> feel that the card is very slow, the test with Bonnie shows 1810KB/sec 
> with CPU load less than 2%. This shows the CPU time is idle, but the 
> load average go up to around 13-14 . This means the dpt_i2o driver is 

The load average includes I/O waiting. So that may well mean you have 14
different processes waiting for the card.

> even faster than this Ultra-160 SCSI i960 RAID with 32MB cache. If this 
> is the case, I will just throw it away and use my onboard Ultra-160 
> connector with software RAID which runs faster(at least). This is almost 
> unusable for even small workgroup server.

The numbers seem extremely poor. There are plenty of abysmally slow
RAID controllers out there but I'd expect bettere than 1810K/second even
on things like old Megaraid cards.

Alan