On Mon, 28 May 2001, Gunther Schadow wrote:
> Anders Magnusson wrote:

> > > So, I could imagine using CI as a link layer for IP just like
> > > Ethernet. Anyone ever tried that or knows otherwise that it
> > > cannot work?
> > >
> > Yes, it is supported in Ultrix and works well (we have used it for
> > several years). It runs about 70Mbit with an MTU of around 8K.
>
> Yes! Cool. Now you do use the star coupler as a hub, don't you?
> Are all 4 CI connectors used for a single VAX-to-starcoupler link?
> If not, and if a crossover VAX-to-VAX link would be physically
> possible, one could build sort of a ring network where every
> VAX connects to two neighbors and all do link-layer bridging.

Without knowing for certain, I'd imagine that the CI setup with the star
coupler isn't a true star topology but rather a full-duplex dual-ring
topology that happens to be run through a star for convenience.  The
4-connector A/B stuff you're mentioning sounds very similar to IBM SSA
storage network solution.

Given the high availability and high performance requirements of a VAX
cluster, it's a logical design.  Under normal circumstances you'd be able
to utilize both loops, but if one loop breaks you can still talk to the
other devices using the remaining loop.  It's a degraded mode, but at
least you're running.

There's a relatively brief RFC for IP over SCSI.
(http://www.alternic.org/rfcs/rfc2100/rfc2143.txt) I've often wanted to
take one of the two SCSI interfaces available on the VS3100s and turn it
into a high speed network interface to run between VS3100s.  Back in the
mid 1990's I rememeber reading about some OSF researchers successfully
implementing an IP over SCSI interconnect using some Adaptec EISA
controllers on an Intel system running under a version of Mach.

I'm not sure at what speed the VS3100 SCSI busses are rated, but at 5MB/s
or 10MB/s this would be a rather impressive interconnect for such an old
system.

-brian.