Re: Vinum raid5 proformance problem

From: Steinar Haug (sthaug_at_nethelp.no)
Date: 02/27/04 

Date: 27 Feb 2004 18:35:19 GMT

[Steve O'Hara-Smith]

| TJ> The raid5 itself outputs a very poor result, a basic ata100 disk
| TJ> delivers a better result :O(
|
| That is not particularly surprising, RAID5 is a cost/reliability
| trade off not a performance enhancer. In particular RAID5 writes are usually
| slow because of the parity calculation requiring several reads. Yours do
| seem to be worse than I would expect though. You should see an improvement
| in read performance with RAID5 - but all your read tests are maxed out on
| CPU usage which is the real limiting factor and a little suspicious. Are
| you sure that you are getting UDMA modes ?

There's another possible problem also. Looking at the performance
results, I see

drive d4 device /dev/ad10s1e
drive d5 device /dev/ad11s1e

Isn't /dev/ad11s1e an ATA *slave* disk? If so, this is part of your
problem, since slave disk operations cannot overlap with operations
on the corresponding master disk. Having a slave disk as part of a
vinum setup is usually a bad idea.

Steinar Haug, Nethelp consulting, sthaug@nethelp.no