On Thu, 13 Feb 2003, Thor Lancelot Simon wrote:

> On Thu, Feb 13, 2003 at 01:58:17PM +0900, Curt Sampson wrote:
>
> > At any rate, if you try an 85xx, let me know how it works. I've used a
> > 7508 in the past and been quite pleased with it, except that the card
> > itself seems unable to process more than about 120 I/Os per second, no
> > matter how many drives you have and how fast the drives are.
>
> That's *extremely* strange.  The humble 6410 in anoncvs.netbsd.org can
> clear hundreds of transactions per second so long as you don't force it
> to seek all over the place -- it's running RAID 1, however.

Well, I was seeking all over the place, because that was the whole point
of the exercise. Looking back at my sequential read tests, though, I see
I was reading (from the raw device) 16K blocks at a rate of about 110
MB/sec, which works out to 7040 reads per second form the point of view
of my little test application (see randread.sourceforge.net). Since it
was the raw device, that would mean that each read was delivered as an
individual request to the controller, right?

So it looks like the seek performance is the real bottleneck here.
Perhaps this baby doesn't really have eight truly independent ATA
controllers on it, and is waiting on seeks on disks before delivering
requests to other disks?

I've still got the system in a box in the corner, if it would really be
worth doing some more testing and/or debugging on the thing.

cjs
-- 
Curt Sampson  <cjs@cynic.net>   +81 90 7737 2974   http://www.netbsd.org
    Don't you know, in this new Dark Age, we're all light.  --XTC