HFISCHER@USC-ECLB.ARPA (05/11/84)
From: Herm Fischer <HFISCHER@USC-ECLB.ARPA>
My company is a defense contractor planning for support of Ada
Programming Environments. While waiting for Ada compilers to really
become available, we are using Unix as the host environment. We
currently need to interconnect three types of devices: Unix host
machines, dumb terminals, IBMPCs with PCDOS (and soon IBMPCs with
UNIX). Our people are quite confused by the barrage of incompatible
announcements, and the following should help establish the status quo.
Three LANs can be attributed to IBM; these are different and
(seemingly) incompatible products:
1) PC Cluster
2) Star-wired, token-ring ("IBM Cabling System")
3) Rumored possible Sytek-derived PC LAN
There are many non-IBM LANs available; we currently have a Sytek
network but remain apprehensive about software support for PCs. Many
other folks recommend the 3-com. This article concentrates on IBM
offerings and nonofferings, because of the focus on the need for
integrated software support at the operating systems level.
PC Cluster
----------
This LAN product (CSMA-CA) was announced last month; analysis of the
manuals indicates that the product is limited to PCDOS use (not
extensible to UNIX), and furthermore, very limited there. It includes
adapters which mount to PCjr side sockets and cards for PCs, each
having a new 8K-byte cluster BIOS program in ROM (at address D0000,
responding to software interrupt 5A), an 8031 micro to control the net
access, and circuitry to attach to a 75 Ohm Coaxial Bus via a passive
BNC Tee.
Software includes both node servers and a disk server.
The node server program loads a DOS interrupt handler which cooperates
with the new BIOS ROM and remains resident thereafter. This node
server allows other cluster-net subscribers to "reach out and touch
somebody" even while they are running their spreadsheets, editors, or
otherwise using the processor. Any node can do directory listings,
file reads, and file writes from any other node on any path,
regardless of what the other node is doing then (so long as PCDOS
remains active).
Headerless messages are accepted by a node server. One sends a
message to a given node and it is queued as expected. The messages
can be only 512 bytes long, and have no subject qualifier field. They
are not compatible with uucp or ARPAnet programs.
Security is weak; any node can identify which of its drives are
readable, writable or neither. If you have a hard disk, it is either
wide open to net-hackers, or inaccessible. If you designate your "a"
drive for net access and mount a different disk, its contents are then
accessible. If you designate a floppy as default and pull out its
message files, things might get messed up. No path control is
provided (as is found with uucp).
The disk server is implemented by having an XT (or clone?) which
establishes areas on its hard disk for three types of volumes: (1) a
public read-only volume, (2) private read/write volumes (one per
network node), and (3) a public "remote system reset" volume. These
volumes appear to remote nodes on the net as additional lettered
drives, for example, to a two-floppied PC, the public drive is C, the
private volume for the node is D, and the remote system reset volume
is E. The order changes for a diskless PCjr; the reset becomes A&B,
the public C, and the private D. The reset volume typically contains
IPL programs (presumably command.com, autoexec's, and other
applications).
The disk server node both serves remote node disk requests, and
continues to be available for local processing; the server station
however has higher priority than the network for disk access. Loading
1-2-3 at the server would hold up net-generated disk requests for a
while. A dbase II help request could have the same effect.
Star-wired, Token-ring
----------------------
This product was announced on May 8th as the "IBM Cabling System".
The announcement details wiring schemes which allow a 150 ohm cable
with two shielded twisted pairs, and four unshielded pairs, to
accommodate present and future networks EXCEPT the Cluster (and maybe
Rumor item #3). A highlight of the announcement is the development
of a new asexual hermaphroditic plug for the coax cable (it is both
and neither male and female). This plug supports the future loop in
that the twin shielded pairs short when the plug is unmated. (The four
pairs are for telephone connections. Will the telephone be an IBM
accessory soon?)
IBM has a demonstration disk with a PC color graphics program which
describes the cables, the token ring, ring bridges, gateways, and
broadband interconnects. The animation of this demo disk is
outstanding. Unusual for IBM, this disk comes with the copyright
warning "Duplication [of the floppy] may be done with no associated
charge, providing that each reproduction is made without alteration."
The disk even depicts a flying saucer irradiating the token ring and
zapping a token, to demonstrate recovery. If you need more, the
flying saucer goes on to vaporize a node after it sends out its
message and before it removes the acknowledgment token, and shows how
the network recovers! For slow learners the animation (token
movement) can be slowed down, and for impatient folks, you can speed
it up. Outstanding, even if there is no token ring hardware for the
PC. (What will the Ethernet folks put on their animated floppies in
retaliation???) Ask you friendly IBM representative for the "Raleigh
Marketing Support Center, Telecommunications Systems, Raleigh N.C.,
Demonstration Diskette, 'IBM Cabling System'", or copy one from a
friend.
Rumored Sytek-derived PC LAN
----------------------------
[I encounter many rumors daily about this "unproduct". The picture
presented is my "compilation" (and optimization) of them. Neither IBM
nor Sytek are sources of any information, and both insist there is no
such product or project.]
Somebody somewhere maybe is building Sytek derived technology LAN
cards for IBM PCs. A large quantity of these cards exist maybe.
Announcement might be at the NCC.
I hypothesize that if that is true, these devices will resemble the
present Sytek network; coax cable (broadband) CSMA-CD, bus style
connectivity (not Star-like as above).
I hope that if IBM builds, buys and markets, or otherwise supports
this product, the direction established by the Cluster will be
followed and improved upon. Maybe the nodes will have PCDOS servers
which run regardless of ongoing software activity, but I hope they
will not lay bare all directories on my hard disk. Maybe message
traffic will be allowed, but I hope it will be usenet/arpanet-like.
Maybe there will be disk server software. I even would love support of
"uux-ing" between sites (e.g., ask another site to do a dbase query
and send me the ASCII-format result file at its convenience).
Finally, I sure hope that Boca Raton (PCDOS folks) has heard of Dallas
(Unix folks); just because they (and other Unixes) are different
shouldn't prevent their co-residency on a bus. An open-architectured
LAN, rather than a closed restrictive product, would have a chance at
survival, especially if its software integration was good.
Recommendations
---------------
There is only one possible recommendation: wait. Do not buy. For
the message is clear, there will be something better. To quote Jerry
Pournelle's favorite utterance, "Real Soon Now".
-------