Arpa address 3487

re:

so a location might have an IMP node for the network, a mainframe ... considered a HOST, another IMP working as a terminal server ... also considered a HOST? The *250 host* arpanet on 1-1-83 then possible was in the range of 100 so IMP network nodes, 100 or so mainframes (considered hosts), possible 100 terminal servers (considered hosts?), and misc fifty other somethings also considered hosts?

the internal network nomenclature was somewhat more mainframe oriented-centric than network nomenclature centric.

you would have a network node that was synonomous with host and mainframe. the mainframe would likely have a half dozen or so "controllers" (in some nomenclature these have been called FEPs, or front-end processors). The controllers could be a mixture of telecommunication controllers and-or terminal controllers. The telcommunication controllers handled a mixture of slow-speed terminal lines and-or telco lines (up to 56kbits-sec). The (3270 dumb) terminal controllers handled coax 3270 terminals that got hundreds of kbits-sec.

Not all of them were the os-360 360 machines ... here is a list of RFCs done by Joel Winett at Lincoln Labs with regard to their arpanet connection with 360-67 running cp67 time-sharing

This post on MTS and UofMich mentions the "60s" 2703 telecommunication controller (used both for terminal controller lines and other telco lines) and Mich building terminal controller from PDP-8 I had been involved in something similar as an undergraduate but using an Interdata-3 instead of a PDP-8

A typical mainframe-hosts configuration (customers, internal network, etc) would have several tens of coax lines to several hundred of coax lines. Through the 80s, a lot of the 3270 dumb terminals on the end of these coax connections changed over to PCs supported with terminal emulation ... some recent comment on terminal emulation paradigm

One of the selling points of T-R starting in the late 80s was reducing the building loading weight by changing the terminal emulation from direct (3270) coax to T-R. The 3270 coax had long coax runs from the machine room to the terminal. As the number of coax cables in these long bundle runs increased ... there were numerous installations where it started to bump up against some structural loading weights. A large datacenter might have tens of mainframes and thousands or tens of thousands of coax cables coming out of the datacenter. A big selling point for T-R was the significant problem that some customers faced with pure structural loading weight of the coax cables ... transitioning from direct point-to-point coax for the terminal emulation to running terminal emulation of T-R. Typical customer strategy was to hang 300 or 400 "terminals" (PCs with terminal emulation) off the same T-R LAN ... all sharing the same bandwidth.

for some drift, when we were out pitching 3-tier architecture in the late 80s ...

we showed a configuration of 100 PCs and workstation. the terminal emulation scenario had typical 100 machines sharing a 16mbit T-R (1-3rd typical configuraiton of 300 machines sharing 16mbit t-r) The 3-tier architecture had multiple enet sections running over the same T-R wiring with high-speed router-backbone for approx. the same cost.

The 16mbit T-R LAN had possibly 8mbit aggregate max effective thruput and each individual T-R LAN adapter peaked at less than 1mbit-sec (there was this chicken&egg reasoning ... if you were only doing terminal emulation you didn't have to design for higher thruput ... and if you didn't have higher thruput you didn't need to do anything other than terminal emulation).

There were a number 16 separate 10mbit enet segments (6-7 machines-segement) going into the high-speed router (that was rated at overall sustained thruput of 200mbit-sec) ... each enet segment getting over 9mbit-sec aggregate and the individual enet adapter boards capable of over 9mbit-sec sustained.

The PC-RT group had earlier done a custom 4m T-R card ... but in the switch-over to 16m T-R card, the group was told that they had to use the same card that was produced for the PC terminal emulation market. This 16m T-R card had lower per card thruput than the earlier PC-RT 4m T-R card.

misc. past posts mentioning the difference of opinion between the workstation group and the terminal emulation forces:

--