[42] in resnet
How to join a net through NET
daemon@ATHENA.MIT.EDU (Jerry Larivee)
Thu May 27 14:19:29 1993
Date: Thu, 27 May 1993 14:16:51 -0400
From: akajerry@watson.ibm.com (Jerry Larivee)
To: ninjo@Athena.MIT.EDU
Cc: resnet@MIT.EDU
In-Reply-To: ninjo@Athena.MIT.EDU's message of Thu, 27 May 93 13:06:38 EDT <9305271706.AA24118@alfredo>
ISDN certainly has its promises, although the last time I
spoke with someone who knew this stuff (a few years ago now) he
pointed out a few rather major problems. I'ld be interested to hear
if they've been solved yet. Here's a refinement of what I was
thinking about at the time along with the problems.
Standard ISDN service, which is what you get at MIT these days
if you have one of the Darth Vader phones, is 2B+D. Two B channels
and 1 D channel. The B channels are 64Kb and the D channel is 16Kb
(actually I thought they were 19.2Kb). It happens that 8bits/sample
at 8KHz (8*8000=64Kb) is what the AD/DA converters in the phone system
work at. All this is uncompressed bandwidth.
But have you noticed that the serial ports in the back of the
ISDN phone set max out at 19.2Kb. I thought that was because they
used the D channel, but I could be wrong. I do know that the 19.2Kb
limit is a pretty hard limit for those phone sets. If you want to get
64Kb then you have to get an ISDN plug in card (I saw one from Hayes
at Summer USENIX '91, but I bet even today they're expensive) for your
computer (the DEC Alpha's and the Sun Sparcstation LX's come with ISDN
built in, although I don't know if it's a full implementation.)
If you want to run packet switched over ISDN you have to use
X.25 which limits you to 56Kb and is really a pain. The best way to
get the maximum throughput and ease of use seems to be two computers
at each end of an ISDN B channel with an ISDN card in each, talking
raw bits. This can be used as a serial line or as a SLIP connection.
So far, so good, the problems come in when you try to scale.
What was pointed out to me is that you can multiplex 24 B channels
over a T1 line, but NYNEX hasn't provided the software for the MIT
ISDN switch that MIT would need to demultiplex them in software and
channel them directly over MITnet. What MIT would have to do is set
up a milking machine style bank of demuxes out of the switch and feed
each B channel into a single serial modules and back into the switch
to connect to the CPTs, much the same as MIT does currently with the
dialup modem pool. This milking machine is very expensive per
simultaneous connection.
Assuming you solve the above problem, which is probably the
biggest, you next have to deal with SLIP. I haven't heard anything
recent from MIT about SLIP, but the last time I heard the problem was
still one of assigning IP addresses. Do you assign them statically or
dynamically? With a properly integrated ISDN solution you could use
the CallID info to map IP addresses to phone numbers. Better yet,
with a special SLIP nameserver domain you could have a limited number
of numeric IP addresses dynamically assigned to static hostname names
which are maped to phone numbers.
The problem seems to be more a question of software than
hardware. MIT might need to buy a specialized CPT or two, but CPT's
seem to be cheap per simultaneous connection. Getting NYNEX to supply
the software for the ISDN switch could be a major problem, I don't
know how responsive their applications development group is, assuming
they have such a group. I also recall there being a problem with
PC/Mac software supporting SLIP properly, and particularly supporting
dynamically assigned numeric IP addresses. Although MIT does have a
class A network, so maybe statically assign numeric IP addresses is
possiblility.
--Jerry
