[Simh] HSC vs UDA/QDA

[Mark Pizzolato]

On Wednesday, March 8, 2017 at 11:50 PM, Johnny Billquist wrote:
> On 2017-03-09 02:26, Paul Koning wrote:
> >
> >> On Mar 8, 2017, at 7:44 PM, Johnny Billquist <bqt at softjar.se> wrote:
> >>
> >> On 2017-03-08 22:15, Bob Supnik wrote:
> >>> The HSC family offered a superset of capabilities compared to the
> >>> UDA50/QDA50. In particular,
> >>>
> >>> - tape as well as disk support (TMSCP as well as MSCP);
> >>> - controller-based disk to tape backups and tape to disk restores;
> >>> - controller-based disk to disk duplication;
> >>> - controller-based volume shadowing (RAID 1).
> >>>
> >>> UDA50/QDA50 did not support tape drives, disk duplication, or
> >>> controller-based volume shadowing.
> >>>
> >>> HSC supported some data caching; UDA50/QDA50 did not.
> >>
> >> Oh. I didnät mean to imply that the HSC was just the same as an UDA. But
> >> the MSCP protocol as such is the same between them. Shadowing, local disk
> >> copying, caching and so on, are just things a controller can do without the
> >> host are even aware of it happening.
> >> But since we're talking emulation, the actual disks now might be doing
> >> even more of that than an HSC ever could. It's not really something that
> >> makes much sense to emulate.
> >
> > I think the significant different for emulation, as opposed for the real
> > hardware, is that CI is a multi-access network (like Ethernet).  All the hosts can
> > see all the disks, and in addition the hosts have peer to peer communication.
> > VAXclusters use both of these things.  You can of course do them via LAVC
> > (same services but over Ethernet).  With CI emulation you get a second way.
> > That enables running clusters with VMS versions predating LAVC, if that is
> > interesting to anyone.
> 
> Yes.
> 
> However, to get that working you will either need to run several emulated
> machines under the same simh instance, or have the CI run outside of the
> simh framework, so that several simh instances can communicate with it.
> While possible, this could turn complicated.
> If you only do CI within CI, with the limit to one machine, then all that is lost,
> and you end up with the same as a local MSCP and TMSCP controller, with
> just a different transport layer you need to implement.

When Matt Burke initially was working on this, I believe that we talked briefly
about extending sim_ether to support packet delivery across IP multicast.  
With that model, all of the systems connected to a particular star coupler 
would use the same multi-cast group.  I think that DEC may have done 
something very similar to this when they implemented LAVC.  Several 
independent Local Area VAX Clusters can certainly coexist on the same 
LAN without interference.

- Mark