Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!ucsd!usc!zaphod.mps.ohio-state.edu!ub!dsinc!bagate!cbmvax!grr
From: grr@cbmvax.commodore.com (George Robbins)
Newsgroups: comp.sys.dec
Subject: Re: 5820 performance ( IO )
Message-ID: <15524@cbmvax.commodore.com>
Date: 1 Nov 90 18:45:38 GMT
References: <1990Oct29.151224.21573@mlb.semi.harris.com>
Reply-To: grr@cbmvax.commodore.com (George Robbins)
Organization: Commodore, West Chester, PA
Lines: 40

In article <1990Oct29.151224.21573@mlb.semi.harris.com> jdr@jrx.mlb.semi.harris.com (Jim Ray) writes:
> Has anyone had experience with 5820's connected to HSC70's ( with
> RA92's )?
> 
> I have been less than impressed so far with the IO performance and would
> like to find out if anyone else has had similar experiences with the 5820.

One thing that it is important to keep in mind is that the DS5800 I/O, especially
the CI/HSC combination, is that it is considerably optimized in the direction of
aggregate thruput, rather than peak single drive transfer rate.

At the moment I have several disk bound jobs running and the effect on interactive
response isn't particularly notable, and I could start a couple backups to HSC
attached TU78's and they would run "full speed".

This is quite different from the kind of I/O perfomance you'd get on a DS3100
or other single SCSI bus system, which might claim several times the single drive
peak transfer rate of the CI/HSC/RAXX combination.

> Also, we have the CI configured as another network and have been transmitting
> data to the other 5820's via this mechanism.  It appears to be 2x slower than
> going over the ethernet connection.

CI networking is an interesting subject.  I suspect the CI is somewhat optimized
towards block transfers and you won't see any decent throuput with the standard
ethernet MTU's, while if you could jack up the MTU and look at something like
NFS wich does block transfers, you might get better results.

Some of the other high-speed networks have similar problems, they just don't
shine when used as pseudo-ethernets.  I dunno whether there's any way to make
the CI really look like an 80M-Bbit/S network in the Ultrix environment or not.

I've heard conflicting stories about whether it's really an effective alternate
inter-CPU backbone link and whether using it as a network substantially
interferes with CI<->HSC performance.

-- 
George Robbins - now working for,     uucp:   {uunet|pyramid|rutgers}!cbmvax!grr
but no way officially representing:   domain: grr@cbmvax.commodore.com
Commodore, Engineering Department     phone:  215-431-9349 (only by moonlite)