Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!rutgers!sri-spam!ames!ucbcad!ucbvax!RVAX.CCIT.ARIZONA.EDU!JMS%uamis
From: JMS%uamis@RVAX.CCIT.ARIZONA.EDU.UUCP
Newsgroups: comp.os.vms
Subject: RE: Terminal servers
Message-ID: <8711111751.AA02222@ucbvax.Berkeley.EDU>
Date: Tue, 10-Nov-87 16:45:00 EST
Article-I.D.: ucbvax.8711111751.AA02222
Posted: Tue Nov 10 16:45:00 1987
Date-Received: Sat, 14-Nov-87 01:40:04 EST
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The ARPA Internet
Lines: 53

Larry Herzlich writes about DECservers and LAVc nodes.
He also includes a quote from Steve Ward.

I believe that both of you are making incorrect assumptions.

There is NO reason to "de-rate" a CPU because you're using
LAT terminals instead of direct terminals.  In fact, if you're
comparing LATs to Programmed I/O devices, you may "up-rate"
your CPU.  If you're comparing LATs to DMA devices, you
may consider them one-for-one swaps.  There is NOT an additional
overhead to dealing with LATs that there isn't in direct
devices.  If you had a DECserver 500 and ALL of it's ports were
connected to a VAX CPU instead of having the same number of
serial ports, you would see a great efficiency even above normal
DMA devices.  For 8-port and 32-port LAT servers, we have
found the response to be as good as, if not better than, DMZ-32s,
Emulex CS21/CS32 combox, DMF-32s and a LOT better than DZ-11s.

Also, your comment about saturating the traffic on the LAN
with LAT traffic is way off base.  We run ALL terminals in this
building on LATs (currently there are about 12 LATs worth of
terminals) and see short term peaks of up to 3% of maximum bandwidth,
but long term traffic is below the 1% level.  Do a little
arithmetic -- 40 users, right.  Input, most users can do a
good average (over short time spans) of as many as 3 characters
a second.  Let's see, that's 40 * 8 * 3 = 960 bits per second. Assume
a 300% overhead for this on LAT terminals, that's about 4000 bits
per second, out of a possible 10,000,000.  For output, let's assume
that your users do lots of types and edits and whatever, and 
are getting as much as 50 characters per second AVERAGE output.
40 users times 50 times 8 is 16000 bits per second, but the overhead
is a lot less; say about an additional 50% to make 24000.  Now, let's
assume that I'm a full order of magnitude wrong on all my numbers
(ie, everyone is ten times as active as I said.  I'm not actually
wrong, but if it'll make it more believable...).  You're still using
only 25% of your Ethernet.  Those out there who know will point out
that the fastest we've ever seen a VAX do terminal input is about
2000 characters a second (about 1.6 MHz, again out of the 10 you
get on an Ethernet).  

Synopsis:  Don't worry.  Put bridges between buildings.  It works
great.  Digital is NOT going to support direct terminal lines 
except as a black-sheep product.

jms

+-------------------------------+
| Joel M Snyder                 |            BITNET: jms@arizmis.BITNET
| Univ of Arizona Dep't of MIS  |          Internet: jms@mrsvax.mis.arizona.edu
| Tucson, Arizona 85721         |   Pseudo-PhoneNET: (602) 621-2748
+-------------------------------+              ICBM: 32 13 N / 110 58 W
(I have gotten into trouble too many times to put any faith in disclaimers)
"There's nothing here that an overdose of Seconal won't cure."