[LINK] Twenty Years of the Net in Australia

[Glen Turner]

On 26/11/09 19:18, Andy Farkas wrote:
>> CSIRO became the first major organisation in Australia to implement a
>> national VoIP service on AARNet’s backbone in November 1999.
>
> Now that's impressive! I wonder what software they used?

Sure. I was one of the two project technical leaders (the other was
Stephen Kingham), so I can answer this.

The aim of the exercise was to reduce long-distance and mobile call
rates to universities. We began this by approaching Telstra and Optus
and asking for cheaper rates (about a 10th of what we were then
paying). They refused and told us that our threat of a toll bypass
VoIP system was weak, since their staff had assured them that it
was not technically possible in a large scale deployment outside of
a demonstration.  Of course, we'd been there before, with the Internet
itself and the telco's telling us it wouldn't work....

We used Cisco 5300 routers with E1 interfaces to interconnect with
existing university PABXs using the QSIG protocol. We configured
the routing on the PABXs to least-cost prefer those routers for
all numbers. We would then send back a congestion response if we
did not service the number, and the PABX would reroute the call
to the next route (such as their usual carrier).  Stephen spent
ages making sure all this worked well.

The VoIP system itself used H.323 as its protocol, and a gatekeeper
kept the routing information of serviced numbers and which 5300
E1 interface they were connected to.

I spent a few weeks trudging through the language schools of unis
recording speech samples of various languages. Vietnamese was of
particular concern to us, because it is multi-tonal. I then simulated
all the codec chocies at all the bandwidths at various packet loss
rates, and sent the resulting CDs back to the language schools for
evaluation. The results were impressive, with the lowest rate codec
giving good results under adverse conditions.

Our backbone was configured to run QoS so that voice calls would
never be faced with IP network congestion. That took me a while
as it was at the very edge of what our 7513 routers were capable
of. Because of router limitations the result was not DiffServ,
but was tailored to what the routers could do. It was however,
DiffServ compatible, so we could easily replace the 7513s when
the time came without losing QoS.

One we had the PABX trials (one from each vendor used by unis), the
QoS, the H.323 routing and the codec confirmed working in small
tests we felt confident we could roll the whole thing out and
from then on it was a straightforward project management exercise
in network deployment, something we had done many times before.
(We call this the point where you come to work in a tie rather
than thongs -- because it's about persuading people rather than
technology).

Stephen and I must have spent months on the phone educating people,
so they would have the knowledge to deploy the system. We ran
handfuls of training courses and seminars. We promoted the thing
at all sorts of university and technology events, as there was
no point getting technical staff interested if their manager
would say "no". The AARNet senior management did a fine job here
too.

The result was that if you called from one university campus to
another your call would go across the Internet as VoIP rather
than use the PSTN. This resulted in a substantial saving for
some institutions, especially CSIRO and the regional universities.
CSIRO was the cornerstone customer, they saw the potential savings,
they believed the technology could work, believe we could deploy
it in a large scale, allowed us to second Stephen, and it worked
out well for them. It was so cheap that we made the price a cent
a call, simply because we didn't want the accounting complexity
of sub-cent rates.

We also connected some E1s to carriers. So if you were at, say,
the University of Adelaide and dialled, say, a Sydney business
your call would go across the university PABX, into the AARNet
VoIP from Adelaide to Sydney, to Optus in Sydney as a local call.
The local call rates were very low, since we were originating so
many calls, less than one-twentieth of what universities had been
paying for local calls, let alone long distance calls. Prices
fell over the years, ending up in the locality of two cents.
You can see what a rort long-distance toll rates are.

The system no longer exists?  Why?  Because as telco contracts
came up for renewal the tender respondents had to beat the
AARNet VoIP system price to get any business.  So the AARNet
system set a ceiling above which prices would not be offered.
This in turn made the AARNet system the most expensive offering.
So we turned it off about a year ago. But it's job was done. We
had massively reduced the price universities paid for voice
telecommunications.

We're now seeing universities deploy VoIP handsets, and we provide
a H.323 meet me point for universities and others that want to
peer those networks for cheap intercalling.  In the coming year
we'll provide a joint H.323/SIP meet me. I doubt we'll ever provide
a dial-out facility from that to other voice carriers, but where
AARNet peers with an ISP and it has a VoIP offering we enable
QoS on that link. This allows universities to accept tenders from
those VoIP carriers, knowing that the result will be a high
quality call.

-- 
  Glen Turner   <http://www.gdt.id.au/~gdt/>