On Sat, 17 Jul 1999 [EMAIL PROTECTED] wrote:
> On Thu, Jul 15, 1999 at 08:29:47PM -0700, Scott Laird wrote:
> >
> >
> > On Thu, 15 Jul 1999, Andrew B. Cramer wrote:
> > >
> > > You would get good performance if you used a multi-channel controler.
> > >
> > > (.02) - Andy
> > >
> >
> > I doubt it -- I got about the same performance with a 3-channel version of
> > the same card. Search dejanews for details, or ask me and I'll resend my
> > last post.
>
> I ran software RAID on three U2W and one UW disk using an Adaptec 2940UW.
>
> Each disk gives around 12 MB/s. But 4 disks * 12 MB/s > 40 MB/s/bus
> So, the bandwidth of the disks exceeded the capacity of the SCSI bus.
>
> Performance on RAID-0 was around 8 MB/s, which is less than I could get
> from using just one of the four disks.
>
> When I put in an U2W controller for the tree U2W disks, and kept the UW
> disk on the UW controller, the capacity of the busses where larger than
> the capacity of the disks on them. Performance rose to something more
> like what I'd expected, 35 MB/s or so. The reason it's not higher is
> probably because of the dated PC it's all crammed into.
>
> My little conclusion is: If the combined bandwidth of the disks on an SCSI
> bus exceeds the bandwidth of the bus, you _may_ see performance that
> is very much lower than the bandwidth of the bus. At least I did.
Strange. In one test I did, I had 6 UW disks on a single UW channel, and
managed to get >37 MB/sec writing with SW RAID 0. I wouldn't recommend
building systems like this -- controllers are a lot cheaper than drives --
but it did work.
Scott
