On Mon, 12 Oct 1998, Benjamin de los Angeles Jr. wrote:
> > If, on the other hand, you want to randomize your spindle load (i.e. ideally
> > have a different process reading or writing to/from each spindle), then you
> > want a large chunk size. Some people also call this overlapped read and
> > write.
>
> You mean you're going to spawn several processes, just to read
> simultaneously from the RAID device. I thought the best way is to use
> RAID software to access the data from the devices concurrently and not
> to worry about creating processes of your own to read on every disk.
I guess I wasn't clear enough. Let me try to put it another way.
If you use a large chunk size, and if you have multiple user processes
running on the system accessing files on disk, it is more likely that each
request will be satisfied by a single disk. If that is true, in an ideal
situation, each of those processes' requests will be satisfied by a
different disk... i.e. N disks can satisfy N different requests
simultaneously.
So (generally speaking) a large chunk size is better for a system doing many
things, while a small chunk size is better for a system doing just one thing.
> > I don't think I understand how the concept of chunk size applies to RAID 1.
> > Since it's supposed to fill one partition and then another, I don't see
> > where chunk size comes into it... could just be me. :-)
For some reason, when I wrote this I was thinking that RAID 1 was linear
mode. Not. As Molnar said, chunk size could be use to control read
balancing behavior on a mirrored (RAID 1) array.
-Andy
Global Auctions
http://www.globalauctions.com
