[Simh] LAVC

[Mark Pizzolato]

On Saturday, March 11, 2017 at 1:16 PM, Johnny Billquist wrote:
> On 2017-03-11 19:41, Mark Pizzolato wrote:
> > On Saturday, March 11, 2017 at 10:06 AM, Johnny Billquist wrote:
> >> On 2017-03-11 17:18, Mark Pizzolato wrote:
> >>> On Saturday, March 11, 2017 at 5:49 AM, Tim Stark wrote:
> >>>> Thanks for many replies. I learned a lot about VAX clustering. I
> >>>> will use multicast Ethernet for clustering.  I have more one
> >>>> question. Is that possible to emulate HSC controllers with LAVC access?
> >>>
> >>> When Matt Burke and I were discussing the CI concept, we picked IP
> >>> Multicast as transport for the very specific reason that all modern
> >>> host systems will have natural user mode programming access to
> >>> implement this functionality.  Attempting to create your own
> >>> Ethernet multicast protocol (non-IP based) will have portability
> >>> complexities on different platforms due to the fact different
> >>> platforms have varying support (and potentially require privilege)
> >>> to present and/or receive raw Ethernet frames to/from a LAN.
> >>
> >> I don't know for sure, but I assume that Tim just meant that he will
> >> let LAVC do as LAVC does, without him doing anything.
> >> And LAVC itself might be using ethernet multicasting.
> >> This is no more an issue than DECnet communication, which also
> >> requires access to the raw ethernet.
> >
> > True.  It could use the existing raw Ethernet access methods, but that
> > approach would have the following issues:
> > 	1) Some platforms require significant privilege (root) to do this.
> 
> Same as DECnet.
> 
> > 	2) He would have to re-invent much of what LAVC does.
> 
> I thought he was talking just about LAVC, not inventing something new.
> 
> > 	3) The same problems that exist today on *nix hosts to provide a
> > mechanism for multiple simulators running on the same host to be able
> > to concurrently communicate with each other will exist.
> >
> > An IP multicast based solution lets the host OS solve much of this.
> 
> Either you or me is missing something. I thought he was just talking about
> using LAVC (and thus raw ethernet) for clustering. Meaning he'll just fire up is
> simh instance, and start running. And that's the end of it.
> 
> Seems like you are thinking he is going to implement something. Not sure
> exactly what. Emulate CI? 

That is where he started this discussion, and from what he had said above, it 
seemed like that was still the plan.

> Possible, but will require a lot more work than just
> transporting the bytes over the network. But, that said, it probably makes
> sense to use IP instead of raw ethernet if you are going to implement this. And
> my reasoning for that don't have as much to do with any network access
> (which you still will want to have for DECnet anyhow), but because CI frames
> are something like 8K in size. So you're going to have to implement some kind
> of complex layer on top of ethernet if you do that. And then you are going to
> get close to what IP/UDP will do for you anyhow, so why reinvent the wheel?

Exactly!

I think I vaguely remember something from my discussion with Matt about their
being some sort of virtual circuit functionality in the set of capabilities the CI/SCS
functionality provides, so leveraging UDP multi-cast along with TCP would be 
natural solutions to both of those problems.

- Mark