Xah Lee's Unixism

Anne & Lynn Wheeler lynn at garlic.com
Thu Sep 9 18:29:16 CEST 2004


Morten Reistad <firstname at lastname.pr1v.n0> writes:
> In 1987 T1's(or E1's in this end of the pond)  were pretty normal; 
> T3's was state of the art. But it is not very difficult to design
> interfaces that shift the data into memory; and 1987'is cumputers
> could handle a few hundred megabit worth of data pipe without too
> much trouble; but you needed direct DMA access, not some of the
> then standard busses or channels.
>
> IBM always designed stellar hardware for such things; what was 
> normally needed was the software. To see what Cisco got away with
> regarding lousy hardware (GS-series) is astonishing. 
>
> There was a large job to be done to handle routing and network
> management issues. BGP4 didn't come out until 1994, nor did 
> a decent OSPF or SNMP. 

even in mid-80s .... t1/e1 ... the only (ibm) support was the really
old 2701 and the special zirpel card in the Series/1 that had been
done for FSD.

in fall 1986, there was a technology project out of la gaude that was
looking at a T1 card for the 37xx ... however, the communication
division wasn't really planning on T1 until at least 1991. They had
done a customer survey. since ibm (mainframe) didn't have any T1
support ... they looked at customers that were using 37xx "fat pipe"
support that allowed ganging of multiple 56kbit into single logical
unit. they plotted the number of ganged 56kbit links that customers
had installed .... 2-56kbit links, 3-56kbit links, 4-56kbit links,
5-56kbit links. However, they found no customers with more than five
gnaged 56kbit links in a single fat-pipe. Based on that they weren't
projecting any (mainframe) T1 useage before 1991.

what they didn't appear to realize was that the (us) tariffs at the
time had cross-over where five or six 56kbit links were about the same
price as a single T1. so what was happening ... customers that hit
five or six 56kbit links ... were making transition directly to T1 and
then using non-IBM hardware to drive the link (which didn't show up on
the communication divisions 37xx high-speed communication
survey). hsdt easily identified at least 200 customers with T1
operation (using non-ibm hardware support) at the time the
communication division wasn't projecting any mainframe T1 support
before 1991.

because of the lack of T1 support (other than the really old 2701 and
the fairly expensive zirple-series/1 offering) ... was one of the
reasons that the NSFNET1 response went with (essentially) a pbx
multiplexor on the point-to-point telco T1 links ... with the actual
computer links running 440kbits/cards with the pc/rt 440kbit/sec cards.

hsdt
http://www.garlic.com/~lynn/subnetwork.html#hsdt

had several full-blown T1 links since the early 80s ... and was
working with a project for a full-blown ISA 16-bit T1 card ... with
some neat crypto tricks.

I think it was supercomputing 1990 (or 1991?) in austin where they
were demo'ing T3 links to offsite locations.

-- 
Anne & Lynn Wheeler | http://www.garlic.com/~lynn/



More information about the Python-list mailing list