Subject: Re: VS3100 SCSI
To: J. Buck Caldwell <buckaroo@liveround.com>
From: Brian Chase <bdc@world.std.com>
List: port-vax
Date: 05/29/2001 15:17:39
On Tue, 29 May 2001, J. Buck Caldwell wrote:

> Say - a question for those HW gurus out there. Is the SCSI port
> (B-channel) on a VS3100/m38 capable of multi-hosting? See, I've got
> these 8 VS3100/m38s, and while I can network them all with 10B2, it's
> not exactly speedy. I know there is a RFC out there for IP over SCSI.
> What are the posibilities of hooking together 8 3100's SCSI-B busses
> and having them communicate this way? I realize this would take some
> kernel-hacking at a minimum.

I think the fundamental difficulty here is that for any given VS3100
system, the host adapter can only be on ID 6 or ID 7.  To avoid
conflicting IDs with the host adapaters on a shared SCSI bus, you could
chain the VAXes together in a linear topology like the following:

  VAX #1  SCSI-A ID 6
          SCSI-B ID 6-.
                      | <-- shared SCSI bus.
  VAX #2  SCSI-A ID 7-'
          SCSI-B ID 7-.
                      |
  VAX #3  SCSI-A ID 6-'
          SCSI-B ID 6-.
                      |
  VAX #4  SCSI-A ID 7-'
          SCSI-B ID 7

And then I guess given that you have an even number of systems, you could
tie the two open ends together to create a ring.  This would cut the
longest hop between two machines in half.

Another interesting option might be something like taking the two busses,
bond them together into a single logical network interface for
communicating between two VS3100s at twice the throughput.  I think the
CPU would get overwhelmed though.

Your options would open up quite a bit if you introduced another type of
system with multiple SCSI adapters that could be configured with an ID
other than 6 or 7.  It could act as a hub.

IP over SCSI would be a fun kernel development project, though I'm not
aware of the practical difficulties with this.  I'd have to learn more
about SCSI.

-brian.