So here is what we did and it works....
Use several drives in a RAID array, and the R/W performance
more than doubles. THEN, we added a SCSI Solid State CACHE
into the mix. It is esentially a huge RAMDISK, with electronics
that has and in and out scsi connector. You can put it between
the physical disks and the SCSI controller, and it has smarts
that cache the blocks and also lookahead, in the RAM.
This also further boosted the heck out of performance.
The SCSI cache come from ATTO Corp. in New York.
They also make regular solid state SCSI disks (RAM+CONTROLLER)
that you can use for a journalling or logging device, to also
further speed up a server. The stuff really kicks butt, and is
also battery backed up.
Hope that helps.
Neal
On Sat, Jun 10, 2000 at 01:34:12PM -0600, Steve Wolfe wrote:
> > Has anyone tried out the new ultra160 scsi?
>
> Nope. ; )
>
> > Since qmail is disk I/O bound I'm interested in hearing
> > or discussing the advantages of a ultra160 machine.
>
> In my opinion, I really don't think it's going to be as large of an
> increase as you think for a single drive, because even though the spec says
> that the *maximum* transfer rate is a huge, huge number, the drive itself
> isn't going to put that much out for more than a very, very short-lived
> burst.
>
> The real, difficult bottleneck to beat is the bandwidth of your SCSI bus -
> which this will certainly raise to huge levels. However, in order to get
> near the limits of the SCSI bus, you're probably going to have to put a good
> number of drives in your array. You'll have to research the specs on the
> drives available to you (actual transfer rate, seek time, capacity, price,
> etc.), and decide how you can get the most bang for your buck. Depending on
> your budget and what's available to you, the 'best' bet can either be a
> small number of larger drives, or a large number of smaller drives. One
> thing is certain, though, the more drives you have in the array (within
> reason, of course), the more total transfer you're going to see, all other
> factors being equal.
>
> steve
>
