dvv@hq.demos.su (Dmitry V. Volodin) (09/18/90)
Have anybody tried this weird combination?
--
------------------------------------+------------------------------
Dmitry V. Volodin |
internet: <dvv@hq.demos.su> |
or <dvv%hq.demos.su@fuug.fi> | Motto coming soon.fmayhar@hermes.ladc.bull.com (Frank Mayhar) (09/18/90)
In article <1990Sep19.112128.2154@swbull.bull.se>, love@swbull.bull.se (Love Feuer) writes: |> B.O.S. does not support UDP/IP over X.25. The reason for this is |> probably that nobody should get the very, very sick idea of running |> NFS over X.25. Untrue. While running NFS over an X.25 link is a singularly bad idea, BOS _does_ support UDP over such links. Specifically, BIND, among others, depends on UDP. We've been running a very similar setup here for several months, and it seems to be working OK. There were some startup problems, and there are still a few things that bite us occassionally, but all in all it works. -- Frank Mayhar fmayhar@hermes.ladc.bull.com (..!{uunet,hacgate}!ladcgw!fmayhar) Bull HN Information Systems Inc. Los Angeles Development Center 5250 W. Century Blvd., LA, CA 90045 Phone: (213) 216-6241
love@swbull.bull.se (Love Feuer) (09/19/90)
dvv@hq.demos.su (Dmitry V. Volodin) writes: >Have anybody tried this weird combination? Yup. We have a running TCP/IP link between Norway and Sweden on two Bull DPX 2/320's. It is not a point-to-point link, but a simulated multi- node network. Bull's own little unix (B.O.S) supports TCP/IP over X.25 with extensions to some existing commands. This is the way we did it: (Simplified) Both machines has the following entries in their /etc/hosts: swbull_ix0 125.0.0.1 medoc_ix0 125.0.0.2 swbull> ifconfig ix0 swbull_ix0 publix x121 <swbulls number> swbull> route xadd bull_ix0 <swbulls number> swbull> route xadd medoc_ix0 <medocs number> medoc> ifconfig ix0 medoc_ix0 public x121 <medocs number> medoc> route xadd bull_ix0 <swbulls number> medoc> route xadd medoc_ix0 <medocs number> If we would want to connect more machines to the same network, we would simply add some more hosts with 'route xadd'. It works rather Ok, but it is *SLOW* due to the 2400 BPS modem on medoc's side. We will soon upgrade to 9600 on medoc and that will probably help the performance a bit. B.O.S. does not support UDP/IP over X.25. The reason for this is probably that nobody should get the very, very sick idea of running NFS over X.25. Mail me if you want the more pervert details of this installation. >------------------------------------+------------------------------ >Dmitry V. Volodin | >internet: <dvv@hq.demos.su> | > or <dvv%hq.demos.su@fuug.fi> | Motto coming soon. Love Feuer Bull Sweden +46-8-7284349 Mail: love@swbull.bull.se -or- {anything major}!sunic!swbull!love
bob@MorningStar.Com (Bob Sutterfield) (09/20/90)
In article <1990Sep19.112128.2154@swbull.bull.se> love@swbull.bull.se (Love Feuer) writes: dvv@hq.demos.su (Dmitry V. Volodin) writes: Have anybody tried this weird combination? Yup. [...and then describes running TCP/IP over X.25, without mentioning SLIP.] Many X.25 implementations (ours, for instance) support IP directly, which (for routing purposes) just sees the virtual circuits as wires. I'm not sure why Mr. Volodin wants to put SLIP in the pile, though. B.O.S. does not support UDP/IP over X.25. The reason for this is probably that nobody should get the very, very sick idea of running NFS over X.25. That's odd. Far less sick things use UDP, like the Network Time Protocol and the Domain Name System. They're both entirely appropriate and necessary to run over long-haul links like X.25. I'd speculate that the reason Bull doesn't support UDP over X.25 involves NFS aesthetics less than the cost of setting up and tearing down virtual circuits for the tiny, "cheap" packets that UDP applications expect to exchange. Ask for a CLNS next time :-)
love@swbull.bull.se (Love Feuer) (09/21/90)
bob@MorningStar.Com (Bob Sutterfield) writes: >That's odd. Far less sick things use UDP, like the Network Time >Protocol and the Domain Name System. They're both entirely >appropriate and necessary to run over long-haul links like X.25. I am very, very sorry. I did not check my sources. It is possible to run UDP/IP based protocols over X.25 under B.O.S., including NFS. Although I was told otherwise by my friends here at Bull. I have remote mounted file systems in Norway to our local machine in Sweden. And it works, even if it is slow due to the 2400 BPS modem at the norwegian side of the network. Sorry again for any headache that my (incorrect) information might have caused. /Love Feuer Bull Sweden +46-8-7284349 love@swbull.bull.se
csg@able.pyramid.com (Carl S. Gutekunst) (09/24/90)
>Have anybody tried this weird combination?
Sure. To place the call, I modified the 4.3BSD slattach(8) utility to open up
an X.25 device, perform call setup, condition the line for data only, and then
change the line discipline to SLIPDISC. A similar trick could be done with a
hardware PAD and a uucico-like dialer. To answer the call, I simply used the
NASA Ames slipsh (SLIP Shell), and invoked it right out of /etc/passwd.
The likely problems are:
- If you are using a PAD, it will have to be able to go 8-bit transparent. In
this case, it will also be imperitive that the answering side emit the right
X.3 parameters before going to the SLIP discipline. This might happen auto-
matically; or you might have to kludge something into slipsh.
- If you are using a host-based interface, either a software PAD or a direct
X.25 connection, it will have to have true line discipline support, and it
will have to correctly set the PAD parameters when the line discipline is
changed. Out of a dozen or so commercial UNIX X.25 implementations that I
am aware of, only Pyramid and Sun do this correctly, although I'd bet that
most of the university ones will, too. I'd guess that Sun's is hard on the
machine because you have to wander in and out of the "if" layer twice; but
I haven't tried it, so I don't really know.
Performance between two Pyramids was very good, although I was using a 72Kbps
link. :-) Over 9600bps, expect interactive response to be sluggish, certainly
more so than over a direct wire or a modem. Some of this will depend on your
ability to tune X.29 behavior, if applicable. (A good X.29 will try to bunch
up characters into full packets.) We also have the Van Jacobson 4.3BSD-tahoe
performance enhancements (as does Sun), which seem to make a big difference on
SLIP.
SLIP over X.25 is not anywhere as weird as it sounds. A proper IP over X.25
implementation (like SNDCF) is a nasty and complex beast. SLIP over X.25 is a
cheap and easy way to get point-to-point IP over X.25, with only slightly
higher overhead. SLIP/X.25 can even run over a PAD, which SNDCF cannot. The
big disadvantage, of course, is that SLIP/X.25 *is* point-to-point, just like
regular SLIP; you have to place a call and keep it there for every X.25
address to which you are likely to connect.
<csg>Patrick.Hayes@cediag.bull.fr (Patrick Hayes) (09/24/90)
In article <BOB.90Sep19164020@volitans.MorningStar.Com> bob@MorningStar.Com (Bob Sutterfield) writes: >In article <1990Sep19.112128.2154@swbull.bull.se> love@swbull.bull.se (Love Feuer) writes: > dvv@hq.demos.su (Dmitry V. Volodin) writes: > Have anybody tried this weird combination? > > Yup. [...and then describes running TCP/IP over X.25, without > mentioning SLIP.] [ and doesn't quite answer the original question either] >Many X.25 implementations (ours, for instance) support IP directly, >which (for routing purposes) just sees the virtual circuits as wires. >I'm not sure why Mr. Volodin wants to put SLIP in the pile, though. This is also what B.O.S. does, more or less. Bull's OSes initially treated IP/X25 (DDN/X25 more or less) as point to point links. B.O.S. now treats the X25 interface as a net through which you must route packets to reach the other machines... > B.O.S. does not support UDP/IP over X.25. > >That's odd. and untrue, as Frank Mayhar has already stated. Now for WHY one would like to use SLIP over an X25 link. I'll use my own situation as an example. I've got a Mac at home, a modem, and the software neccesary to use SLIP. I'd like to be able to connect to the machines at work from time to time, but all the access points are on TRANSPAC (the french national X25 carrier). Right now I use a terminal emulation program to log on, but I'd rather have an IP link. The major problem seems to be that B.O.S.'s X3/X29 interface uses one line-discipline, and SLIP uses another. Trying to use both blows my connection. Now I know that this is ugly, but wouldn't it be nice if it worked? Pat -- +-------------------------------+-----------------------------------------+ | Patrick Hayes | EMail : Patrick.Hayes@cediag.bull.fr | | BULL CEDIAG | or hayes@bull.fr | | 68, Route de Versailles | or ...!mcvax!inria!bullfr!hayes | | F-78430 Louveciennes FRANCE | Tel : (33 1) 39 02 49 55 | +-------------------------------+-----------------------------------------+
csg@pyramid.pyramid.com (Carl S. Gutekunst) (09/25/90)
>The major problem seems to be that B.O.S.'s X3/X29 interface uses one line- >discipline, and SLIP uses another. Trying to use both blows my connection. Bingo. You're probably stuck. A *good* X.29 implementation will operate as a real tty device driver, and hence be able to use any line discipline. This is needed not only for SLIP, but also for xt and sxt drivers. Alas, many vendors, including Bull apparently, implement X.29 *as* a line discipline; so trying to switch to SLIP kills X.29. The other blunderous and common way to implement X.29 is to not hook into the tty driver at all, or only minimally. Then a request to change the line discipline is simply ignored. (Note that we're talking all Berkeley/V7 style tty devices here. Under Streams all the rules change. Generally, it's easier to get it right with Streams.) [Enter Flame Mode: My big gripe with virtually everyone who is implementing CCITT and ISO protocols on UNIX is that they insist on forcing new and unusual paradigms on the users, instead of implementing the protocols in such a way that they fit well into UNIX paradigms. The present implementations of X.400 are especially extreme examples, seeming to have been designed in a vacuum.] <csg>
sjg@sun0.melb.bull.oz.au (Simon J. Gerraty) (09/27/90)
In article <128135@pyramid.pyramid.com>, csg@pyramid (Carl S. Gutekunst) writes: >>The major problem seems to be that B.O.S.'s X3/X29 interface uses one line- >>discipline, and SLIP uses another. Trying to use both blows my connection. I have obviously caught the tail end of this chain... Could some one please explain why anyone is trying to run SLIP on top of the X3/X29 service ? In article <1990Sep19.112128.2154@swbull.bull.se> love@swbull.bull.se (Love Feuer) writes: >B.O.S. does not support UDP/IP over X.25. The reason for this is >probably that nobody should get the very, very sick idea of running >NFS over X.25. To my knowledge the DPX/2 supports IP over X.25 without fiddling about with SLIP. I have not seen any references that indicated that UDP/IP should not work. [But of course I haven't tried it myself :-)] >[Enter Flame Mode: My big gripe with virtually everyone who is implementing >CCITT and ISO protocols on UNIX is that they insist on forcing new and unusual >paradigms on the users, instead of implementing the protocols in such a way >that they fit well into UNIX paradigms. The present implementations of X.400 >are especially extreme examples, seeming to have been designed in a vacuum.] Perhaps its because X.400 was designed in a vacuum ? :-) -- Simon J. Gerraty <sjg@sun0.melb.bull.oz.au> #include <disclaimer> /* imagine something *very* witty here */