[comp.dcom.lans] Request For Opinions: Terminal Servers vs. Dataswitches

dd@ariel.unm.edu (03/31/89)

At the University of New Mexico, we invested in campus-wide LAN
technology back in the dark ages, when all terminal servers
spoke a proprietary protocol.  We had a crappy dataswitch, that
for a number of reasons, could not have been upgraded.  We
migrated to a terminal server environment, therefore.  We
actually still have some of those original terminal servers (is
there any market out there for low-serial-number Ungermann Bass
NIU-2As?? ;-).

I was hoping that by sticking my head in the sand far enough, we
would see all RS232-based things on campus disappear - no such
luck!  At this time, we are reevaluating the provision of RS232
services on campus.  Among the viable options are:

[1] TCP/IP-based terminal servers.  I have evaluated some of
    these, and I think I have found the best bet for us.  There
    are problems with this idea, though:  I have to find an
    inexpensive (read "free") implementation of TCP/IP for VMS
    systems (and it would be best if it worked with VMS 5.x), or
    I have to front-end those systems with a TCP/IP terminal
    server (the "milking machine" configuration).  And the same
    for other systems which are not TCP/IP capable, although
    these present much less of a problem.

[2] A dataswitch.  My first reaction was to grimace and shudder,
    but the economics of data switches aren't real bad, and you
    can avoid a lot of the normal problems of LOTS of wire going
    EVERYWHERE through the intelligent use of multiplexers.
    I also get to avoid all the problems of protocol - I just
    have to put a lot of RS232 on every machine in sight :-{.
    Obviously, there are still some reasons not to do it, but
    I guess it isn't as ugly as I originally thought it was.

I would sure be interested in hearing from people who feel
strongly either way - and please tell me why.  As always, I will
provide a summary to anyone interested.  Thanks!

Don Doerner				dd@ariel.unm.edu
University of New Mexico CIRT
2701 Campus Blvd, NE
Albuquerque, NM, 87131			(505) 277-8036

root@helios.toronto.edu (Operator) (04/01/89)

I'd go with the servers. There is a cheap (nearly free) version of TCP/IP
available for VMS. I've forgotten the details, but it's the Carnegie-Mellon
implementation. You can probably get more details about ordering, etc.
off the comp.os.vms list (AKA info-vax on BITNET). The big advantage is
that an awful lot of new systems, in particular *everything* using any
flavour of UNIX, will already come with TCP/IP, and with the servers
you can immediately connect over the network without having to add extra
hardware to the new system. If you stick to data switches, everything you 
ever install from now on (and all the existing systems) will have to have 
serial lines put into them. If you are talking about uVAXes or VME-based 
systems, you can use up a lot of valuable slots this way. We have found
that there are just too many other things we want those slots for. Memory
and disk controllers, for example. If you're going to be putting the
Ethernet in for networking anyway, you might as well take advantage of it.

Other advantages of servers are that a single person can connect to a
number of different machines at the same time, and easily move between
connections with only a few keystrokes. The serial switches I have seen
sometimes allow this, but it isn't nearly as easy. TCP servers, like switches,
can be connected to devices like modems and accessed from the CPU for
dialout. Also, when you need to increase the number of lines available
to users, with servers you just plug another one in (in areas of high
concentration you can use a DELNI or similar to reduce Ethernet tapping).
If your Ethernet is thickwire or you have a DELNI, you don't need to
disrupt the rest of the network to install a new server, whereas putting
new boards into a serial switch system often involves powering down at
least a portion of the other lines. You can also put the server close to
the area where the lines are needed, thus reducing wiring, which can't
always be done with switches (that's assuming that your Ethernet isn't
just contained in the machine room; most aren't).

We gave up on serial lines a long time ago. We have been using DECservers
for access to our VAxes and have recently begun to install TCP servers.
Eventually we hope to have all our DECservers replaced with another brand
capable of doing both LAT and TCP. There are several of this type of server
available now. We have over 200 lines available to users, with direct access 
to either 8 VAXes or every TCP/IP machine on campus (a couple of hundred). I'd
hate to think what the wiring would have looked like to set that up with
serial switches.

Standard disclaimer: this is me expressing my opinions, although obviously
in this case they coincide with those of my bosses, who pay for the equipment.
-- 
 Ruth Milner          UUCP - {uunet,pyramid}!utai!helios.physics!sysruth
 Systems Manager      BITNET - sysruth@utorphys
 U. of Toronto        INTERNET - sysruth@helios.physics.utoronto.ca
  Physics/Astronomy/CITA Computing Consortium