Re: TCP Suite, and other networking.

On Fri, 4 Aug 1995, Andrew Valencia wrote:

> [jsmith@red-branch.MIT.EDU writes:]
>
> All we have is KA9Q. I have a /inet server mostly married with it, which
> permits programs to open, say, /inet/tcp/25, and take the SMTP port for the
> box.
>
> >If there are people working on this I would be more than happy to
> >coordinate things with them so that we do not duplicate any efforts.
> >Stuff that I will be needing to support in the near future are the
> >following: RIP, RIPII, SNMP, SNMPII, SLIP, CSLIP, PPP, HDLC, and a couple
> >of protocols for ISDN work.
>
> I don't know if the version of KA9Q I ported is up to the task. In any
> case, it should provide some shortcuts for the port of some other networking
> code.

Okay, it looks like I will be building most of it up then. Code examples
are always nice however, which is what I suspect most of KA9Q will turn
into. As long as I'm not stepping on any toes. Let me take a look at a
few things then I'll post a summary of how I think I should organize
things for people's comments.
 
> >The second major project would be porting VSTa to some Alpha based
> >computers that I have designed and built. Current OS work on them
> >includes Linux, NetBSD, FreeBSD, and HURD. Some of the machine models
> >are indeed SMP machines. Has someone been working on multiprocessing
> >primitives, or is that still at the stage of just 'holding' places for
> >them in the code? Again I would prefer to coordinate things so that
> >duplicated work is not done.
>
> I have a dream that my annual bonus (coming out in a couple months) will be
> big enough that I can "sneak in" the purchase of a dual processor Pentium
> system. Until then, the only sanity checks which have been applied to the
> SMP support is kernel preemption, and my experience doing SMP at Sequent.

Any of the available low-end motherboards out these days actually any good?
Not too far make most of them were pretty 'iffy'. There are some nice
chipsets for building SMP server type machines out that I have worked with,
perhaps putting together a 2 or 4 cpu pentium system would be something a
few people would be interested in? Motherboard construction costs would be
the biggest cost factor, if there are only a few people interested, if
there are say 10-20 the motherboard fab costs should be about 150-220$
apiece. The chipset I have in mind runs in around 230$ per cpu for
everything, that plus cpu costs and memory. The cpu's hook together,
with or without a L2 cache, via a high speed systems bus,
~560Megabytes/sec. Put a PCI bus on that and a PCI to ISA bridge chip
and you would have a reasonable system, for not too much money or work.
 
> >Somewhere down the line when I have the time, I am also interested in
> >experimenting with VSTa as the core kernel for a MPP system. I also want
> >to work with VSTa in a distributed environment.
>
> This was the primary motivation behind getting /inet up and running. I
> wanted to map VSTa messages over a box-to-box protocol, and thus be able to
> access an arbitrary VSTa server from any other box. This would provide the
> starting point for clustering.
>
> Regards,
> Andy
>

Well hopefully in a few months, I can have the working test-bed of
machines that I mentioned earlier up and on the internet for people to
look and see, and help work on if they so desire, or even play.

At this time it stands at this:

  4 Io Station machines (lower end Alpha single cpu models)
  2 Europa Station machines (mid end single cpu)
  4 Adrastea Stations (high end single cpu)
  1 Europa Station (mid-high end dual cpu alpha, server class)
  1 Callisto Station (High end 4 cpu alpha, server class)
  1 Metis 64 PE MPP system (for starters it will only have 32 PEs)

The Io stations will be used as front ends for the Metis system, with the
rest of the machines hooked into an ATM network(locally) for a compute
cluster. All of these machines can be used for experimenting with VSTa
or what not if needed, I have others that are currently being used for
Linux, NetBSD, FreeBSD development. The reason for the few months wait
is both network connections and for the next run of system boards.

At first the internet link will probably only be 128Kbs, I am working on
alternatives. I may also be able to work out an arrangement with the
Local university to hook my network up to thiers (for the development
stuff) in exchange for them using some of the machines for thier own stuff.

Hopefully I can get interested people to help out some. I have no
problems with partitioning the netowork either, into pieces, some for
VSTa, some of other things. Perhaps some for distributed some for
regular work.

Jonathan
Received on Sat Aug 5 11:50:25 1995