Last modified: Wed May 20 1998
The story
It's a good idea to open up a new browser window to be able to look at the
map
at the same time as you read.
In the beginning...
In August 1996 Michael Strömberg, Örjan Ljungqvist and I was
asked if we could setup a multipoint-to-multipont videoconferencing network
between IT/KTH and EE/SU for a graduate course named
2G1309 Telecom Systems Design Seminar. And we did it!
The original Plan was to connect 5 machines in Kista with 5 machines in
Stanford via a link consisting of 5xT1, approx 7.5 Mbps. The link was
already in place, so all we needed was connectivity in the end-sites.
Considerations
Due to the existing setup and requests from the course-examiners we were
faced with the following constraints at startup:
- None of the link end-locations currently available could easily be
out-fitted with 5 more host machines.
- How to ensure full link utilisation (7.5 Mbps) to each machine?
- The examiner on the Stanford side wanted to use PC's and a
videoconferencing tool called Communique! from InSoft, and in IT
the examiner chose to use SUN Ultra1 and was open for any choice of
videoconferencing tool. Fortunately Communique! was available also for
Unix, although quite expensive.
- An extremely tight deadline, so we had no time to finely tune the
system.
We also had to take care of a lot of administrative stuff besides. For
those interested in history, a bunch of material is available in the
directory /afs/it.kth.se/misc/projects/stanford/Document/TSDS_1996.
Our solution
IT side
To solve the above we did the following:
- We needed to extend the link to some rooms accessible to students.
The Link ends in magda-gw. We extended the link to inner-gw using two
simplex 10BaseT Ethernets to get approx 8Mbps avg. gain.
This was never considered to be nothing but a temporary solution.
The fact that it is still in use instead of being replaced by a better
solution is nothing less than amazing.
- We needed 5 slots for a dedicated Ethernet between the router and
each of the host machines (of which ronin was one). There was (is?)
not enough slots left in magda-gw for 5 more interfaces. This was
another reason to extend the link to inner-gw.
We also encountered the following problems while working:
- Multicast was still not working internally in IT, even though it
worked from magda-gw and most of the Stanford side. So we had no
multicast connection between inner-gw and magda-gw (and we still
don't, but for other reasons :).
- The Communique! version used by the Stanford side PC's were not
the latest version and did not support multicast. The Communique!
version on the IT side almost (!) supported multicast, although we
never had enough time to try it out fully.
This combined with the less than intuitive user-interface and the crappy
silence-suppression function of Communique! made the first real, non-test
run a total fiasco.
So what happened?
The multi-host to multi-host scenario was abandoned, and instead we
flew over a SGI to the Stanford side and ran point-to-point between
one of the original machines and the machine in Stanford.
And this is where we are today.
Maintained by
Tobias Öbrink