skip@gatech.UUCP (11/06/83)
The AT&T network will be offered as being tightly coupled with the System 85,
so it is unlikely that it will not be marketed directly to the end users,
especially since some of our account team is already trying to sell us one
(ie. they are already marketing to end users). The name, as I've heard it,
will simply be Wide-band Premise Network (WPN). It might be sold to the local
BOC's as 3Bnet, since 3B means something to them, but not to the end users.
I've recently heard quite a bit about the WPN which no one has me asked not to
repeat, so if anyone wants any info, please feel free to send mail.
-- skip addison {akgua,allegra,rlgvax,emory,msdc}!gatech!skipskip@gatech.UUCP (11/08/83)
The local area network AT&T has developed and will sell after the first of
the year is quite an interesting one. It is based on a star configuration.
Terminals, hosts, micros, etc. are attached to a board. The board can be
either on a concentrator or on the central switch. Optical fibers connect
the concentrators to the central switch. In both locations the fiber
interface is simply another board on the backplane. The design is extremely
modular. As new devices (Unibus adapters, SNA gateways, X.25 gateways, etc.)
are developed, they are simply plugged into the backplane.
The backplane is simply a very short bus. It operates such that the round-
trip propogation time from one end of the bus to the other and back is less
than one bit time. AT&T will first offer it at 8 Mbps, I believe, but they
have had it working in the labs at 40 Mbps and hope to get it to 320 Mbps.
[ Sheesh!] The contention for the bus is handled by a combination
priority bit / source address scheme which is made possible by the fact that
the bus is less than one bit delay long. When a board has something to send,
it waits for the beginning of the next packet slot (packets are fixed
lengths) and then begins transmitting priority bit(s), source address, then
data onto the contention bus. Open collector outputs are used to put the
bits on the bus, so if the board is transmitting a "1", a "1" will appear
on the bus, regardless of what the other boards may be transmitting. If
the board is transmitting a "0", it will only appear to be a zero if no board
is transmitting a 1. Whenever the board detects that it is trying to
transmit a "0" and the bus has a "1" on it, the board realizes that it has
lost in contending for the bus, and it tries again on the next time slot.
Since all boards have different source addresses, the board with the highest
address wins, unless the lower address board has a priority setting which is
higher. Several priority schemes have been under consideration by Bell Labs
and I don't know which one they've chosen. One scheme would allow for
five priority groups for asynchronous devices while guaranteeing that
synchronous devices could communicate with each other every <xx> milliseconds.
That scheme uses 3 priority bits.
When the packet is put on the concentrator backplane it is picked up by the
Remote Fiber Interface (RFI) and sent to the packet switch where the
RFI there puts the packet on the backplane. The address translator picks
up the packet, does the source-destination translation (I don't know how
a call is set up), and puts the packet onto the broadcast bus. There are
two buses, the contention bus and the broadcast bus. A clock module is
provided to let everyone know when the beginning of the time slots are.
A concentrator and packet switch are very much the same, differing only in
the operation on the address translator.
This info should be more reliable than my info regarding the AT&T micros
since I've heard most of this from at least 3 or 4 sources. The guy who
told me about the AT&T micros also told me that the backplane of the
packet switch was fiber -- this was disputed by everyone else I talked to --
I should have suspected something was wrong with his info about the micros.
-- skip addison {akgua,allegra,rlgvax,emory}!gatech!skip