TCP-IP%brl@brl-bmd.UUCP (TCP-IP@brl) (08/26/83)
TCP/IP Digest Thursday, 25 Aug 1983 Volume 2 : Issue 13
Today's Topics:
Administrivia && Connecting IBM Mainframes to Foreign Devices
TCP/IP from IBM && Looking for TCP/IP to SNA Protocol Gateway
BBN TCP has Retransmit-overtaking-Persist Bug
Comments on the Parsing of HOSTS.TXT
2 more implementations of TCP/IP for VMS
Seeking Portable TCP/IP in Pascal or ADA
Information on Omninet hardware
----------------------------------------------------------------------
TCP/IP Digest --- The InterNet Digest
LIMITED DISTRIBUTION
For Research Use Only --- Not for Public Distribution
----------------------------------------------------------------------
From: Mike Muuss <Mike@BRL>
Subject: Administrivia
This is the first digest in quite some time, caused by overwork, and
slow rate of submissions. There are quite a few interesting topics
still left to be discussed, though!
One important item which I would like to bring to your attention:
For some time now, the NIC has published a much smaller mailing list, on
a direct re-transmission basis, which carries the (perhaps unfortunate)
name <TCP-IP @ SRI-NIC>. A great deal of mail directed to that list
would be profitably displayed in this Digest as well.
The most interesting transmissions to the NIC list I have requested
permission from the authors to reprint, and (so far) have always been
granted permission. However, this is a tremendous administrative burden
on me, and it is no longer my desire to continue with this strategy.
Furthermore, a great many people have urged me to automaticly publish
all traffic on that list in this digest, as they wish to "keep informed",
a request which seems quite justified.
Based on a fair quantity of writing with various people, and a good deal
of contemplation, I have decided to begin printing excerpts from the
NIC TCP-IP (direct) mailing list in the TCP-IP Digest, so that readers
of the Digest can stay informed. Please be aware of the fact that messages
which get sent to the NIC list *might* be published in the TCP-IP Digest.
I will, of course, attempt to be discrete, and will not reprint messages
which "obviously" should not receive wider distribution than they already
got. Good examples are crassly commercial comments, and random flaming.
But, I feel compelled to broadcast solid, technical discussion out to the
widest possible audience, to attempt to increase the understanding and
acceptance of TCP/IP, and to sensitize computer people to the demands
and benefits of inter-operable networks. The subscribers to the NIC list
have been notified of this new policy.
Please direct your comments on this topic to TCP-IP-REQUEST @ BRL,
and I will summarize the response.
Best,
-Mike Muuss, Moderator
------------------------------
Date: Thursday, 30 Jun 1983 09:29-PDT
To: tcp-ip@brl, local-nets@mc
Subject: Connecting IBM mainframes to foreign devices
From: imagen!cpr%Shasta@su-score
There are currently two basic routes to go to connect your IBM mainframe
to special devices: the 4300-series DACU, and the Auscom general
channel-to-Qbus. ACC also makes an IBM channel attachment for Ethernet
which emulates a 327x cluster controller, with individual Ethernet
stations corresponding to 3278 tubes or 328x printers. I haven't been
able to find out more from ACC, and it sounds like a special-purpose
solution, so I won't go into it.
For the 4300 series mainframes, IBM is trying very hard to support OEMs
and customers with special device-connection requirements, using what
they call a DACU (Device Attachment Control Unit), which is basically a
fast buffer between a block multiplexor channel on a 4300 and a true
Unibus, with an IBM PC (personal computer) as the controlling device.
The cost (no OEM pricing yet) is $13,500, without the PC (which only
requires a minimal configuration, of a cost around $2500). The
contact, Bill Denson (Information Systems Group, Atlanta,
404-238-4710), is extremely helpful and informative. Seems that IBM
has finally realized there's money in attaching foreign devices to
their mainframes.
Auscom is a company in Austin, Texas, whose sole business for years
has been making IBM channel attachment devices. The kernel of their
interface is a 3-quad-board Qbus set, which they sell alone for about
$8k, or packaged in an LSI-11 system with software to drive a whole
slew of devices, for about $20k. (Don't believe the prices; talk to
them.) The contact is Linda Lewis, 512-836-8080. I'm quite impressed
with them; they appear to be the only company making this their entire
business, and their customer list is top-notch. For example, they have
a standard channel-to-Ethernet (with simple DoD IP-based protocol),
emulating an IBM tape drive or line printer, etc. (They use Interlan
Ethernet Qbus interfaces.)
--Chris Ryland, IMAGEN Corporation
------------------------------
Date: Mon 8 Aug 83 17:57:04-PDT
From: Suzanne Johnson <JOHNSON@SUMEX-AIM.ARPA>
Subject: IBM TCP/IP
To: tcp-ip@BRL.ARPA
My understanding is that although trial versions of IBM TCP/IP are becoming
available, IBM has not worked out any method for making a product out of
this software. That means that they are not planning a way for a site to
arrange for support service other than through (I believe) an expensive
contract situation with their Federal Systems Division.
It is therefore important that if you are interested in this software, that
you have your local IBM rep call the IBM Special Products Group in Chicago
and say that they have a customer site interested in a supported version
of the software. If the SPG gets 10 or so of these calls, they begin to
believe that they need to establish a product related way to handle the
software.
If only tcp/ip were a bit better known outside of DoD related communities,
it might occur to some of the organizations which are implementing internal
LANs, and scratching their heads over what protocol to use, that tcp/ip is
a natural to consider in this respect. Especially since many LAN's contain
many of the mainframe/os combinations currently supported by tcp/ip
implementations.
Suzanne Johnson
------------------------------
Date: 10 August 1983 09:30 edt
From: Vinograd.Multics@mit-multics
Subject: TCP/IP-SNA Gateway
To: TCP-IP@brl
I am looking for any information on a TCP/IP to SNA gateway. What I have
in mind is the ability to telnet/FTP to any host on an SNA net, given a
physical connection to one host on the SNA net. The reverse access from
the SNA net is equally important.
SMTP support would be useful, but is not a requirement.
Any pointers or rumors of such a capability would be most helpful.
Thanks - Dave
------------------------------
Date: 6 Jul 1983 10:53:51 EDT (Wednesday)
From: Dennis Rockwell <drockwel@bbn-vax>
Subject: retransmit overtaking persist bug
To: tcp-ip@brl-vgr, tcp-ip@nic, bbn-tcp@bbn-vax
There is a bug in the BBN TCP timer code which causes connections
with large delays to hang. The symptom is that the sender will
continually send single-octet packets which are one octet past
the receiver's advertised window. The cause is that the persist
timer (used for probing closed windows) was fixed, which the
retransmit timer is adaptive (variable). When the persist timer
goes off, it resets the retransmit timer. Thus, when the retransmit
timer exceeds the persist timer, you hang.
The fix is to replace the token T_PERS in tcp_procs.c (about line 250)
with tp->t_xmtime*2. This is the only instance of T_PERS except for
its definition (which you can delete if you wish). This guarantees
that the persist timer is always greater than the retransmit timer.
If you know of any system running the BBN software that doesn't receive
one of these mailing lists, please inform either them or me.
Sorry to send this out to such a wide audience, but this bug will
bite more systems as the Internet grows.
------------------------------
Date: 12 Aug 83 15:43:01 BST (Fri)
From: Steve Kille <steve@ucl-cs>
To: tcp-ip@brl.arpa
cc: robert@ucl-cs
Subject: Parsing of hosts.txt
We have found a problem which some sites are having with the UCL-CS
hosts.txt entry. It appeared in the BBN UNIX software, but
this may well not be the only guilty system.
1. Some SMTP sites check the name of a caller against the callers address,
thus if you use a multi-homed host for mail under a single name
it is useful to put all the addresses in the NIC hosts table.
2. UCL has the facility to route mail over SATNET or IPSS so we use
two addresses for UCL-CS (128.16.9.3 the main address and 14.0.0.9)
3. BBN software for SMTP compiles a mail host table from the NIC
tables, it sorts any multiple addreses against the host name. Thus
HOST : 128.16.9.3, 14.0.0.9 : UCL-CS,UCL :: LOGICAL-HOST : IP,TCP/SMTP :
becomes
UCL-CS,UCL : 14.0.0.9 : 128.16.9.3,
Thats OK, but the mailer only ever uses the first address.
The whole point of arranging the addresses in the original
table was to cause mailers to try the first address first.
4. Unless some activity at UCL has opened the IPSS tunnel all
attempts to reach 14.0.0.9 will fail; because of time zone
differences this is quite likely.
Thus it looks as though UCL is hardly ever up, and when I
complain to people about their mailer, they complain ours is
never up.
There seems to be an assumption, valid or otherwise, that all
Internet paths are either up or down, but never
UNI-DIRECTIONAL!
Robert Cole + Steve Kille
------------------------------
Date: Wednesday, 27 Jul 1983 09:35-PDT
To: tcp-ip@BRL-BMD.ARPA
Subject: Re: TCP/IP for VMS
From: Chris Kent <decwrl!kent%Shasta@su-score>
Kashtan's stuff works and seems to be available from the Wollongong
Group. It's full 4.1c networking code.
The people at Rice that did the Phoenix Unix under VMS emulator are
also reported to have the Berkeley TCP/IP running under their system,
but I don't know details.
Cheers,
chris
------------------------------
Date: 6 Aug 1983 1121-PDT
From: LYONS@usc-isi
Subject: TCP-IP IN HOL
To: TCP-IP@brl
cc: LYONS@usc-isi, LYONS@dca-ems
I AM INTERESTED IN KNOWING OF HIGH ORDER LANGUAGE IMPLEMENTATIONS
OF TCP AND IP WHICH ARE PORTABLE, ESPECIALLY IMPLEMENTATIONS
IN PASCAL OR ADA. DO YOU KNOW OF ANY?
REGARDS, BOB LYONS/DCEC
[ I believe that the CSNET TCP/IP implementation for the IBM was written
mostly in PASCAL. There is also a commercial version in PASCAL, for
Cybers and other mainframes, mentioned in earlier Digests. -Mike ]
------------------------------
[ This message is reprinted with permission. -Mike ]
Date: Wed 27 Jul 83 10:17:39-PDT
From: Chris Ryland <g.Ryland@SU-SCORE.ARPA>
Subject: Re: request for Omninet vs VAXes info
To: local-nets@MIT-MC.ARPA
I've been looking into Omninet lately for other reasons, and, as far as
I can tell, there isn't much activity with interconnection to VAXes, or,
for that matter, with other networks. Omninet DOES have an XNS packet
encapsulation protocol, which they and Xerox agreed to (it's published
in the Omninet protocol handbook). There is, I believe, a Unibus
Omninet board just announced or to be announced, though I can't find the
information right now. With that, I suppose you could write a VMS
driver for Omninet.
For people's information, Omninet seems to be the dominant "cheapo" LAN
for the micro world right now (they claim over 20,000 networks, of average
size 4 (stations)). It's a 1mb twisted-pair RS422 network, using two
proprietary chips (they sell) and a Motorola 6801 (with their custom code
burned in) to accomplish the link-level, and little of the transport level.
Thus, for the IBM PC, their board is very simple, and low-cost (about $300,
I believe), as well as reasonably efficient, as they DMA from the network
to waiting buffers in the CPU.
It's really a wonderful network from the point of view of cabling: the
"transceivers" cost $10, and can be wired by anyone with a screwdriver.
1mb isn't bad for a small cluster of workstations. There are some limitations
on the number and type of connections a given workstation can have open: only
one "remote disk" connection at a time is allowed, and only three more other
connections of non-remote-disk type are allowed simultaneously.
/Chris Ryland
------------------------------
END OF TCP-IP DIGEST
********************