Subject: Re: RAIDframe queuing policies other than FIFO?
To: Blair Sadewitz <blair.sadewitz@gmail.com>
From: Greg Oster <oster@cs.usask.ca>
List: tech-kern
Date: 08/13/2007 12:05:50
"Blair Sadewitz" writes:
> Has anyone compared the performance of the other RAIDframe queue
> strategies? I'm especially interested in reasons to choose one based
> upon different workloads. There is precious little documentation on
> this other than comments in the source and certain academic papers
> which are a bit too wide (or narrow) in scope to be of general
> usefulness to NetBSD users.
>
> Additionally, would it make sense to choose a bufq strategy in the
> kernel based upon the RAIDframe queueing strategy, or vice versa? For
> example, using BUFQ_PRIOCSCAN with RAIDframe's sstf, or using disksort
> with cscan? What about fcfs with sstf? Inquiring minds want to know!
> ;)
Inquiring minds might also want to know that there is no support
for alternate RAIDframe queuing strategies in the component label
stuff -- the use of fifo is basically hard-coded for anything
using 'autoconfig'.... So you'll need to do a 'raidctl -c' and not
use autoconfig to get the alternate queue type...
I'd be interested in seeing any benchmarks on this as well....
Later...
Greg Oster