Mike Lawrie
Manager:UNINET
This paper looks at some of the issues in the relationship between Uninet and School Networks. In particular, it gives some insight to the costs of running Uninet, and how these costs are passed on to the sites that connect to Uninet.
Uninet is the research and academic network of the country. It provides inter-connection among the 21 universities, 14 technikons and 15 other institutions, and connects them to the Internet. It is not a donor agency, it does not do this for free, and it has to pay out real money for the underlying services on which the network is built.
There are some fundamental truths about networking, and schools are not exempted from these truths. Some of these are:-
By way of illustration, the budget projections for Uninet that were made in July 1997 for the 1997 calendar year show that the Uninet backbone will cost R2.0M, and the USA bandwidth will cost R6.9M.
The way that things have evolved in the few months since that projection, it is likely that the USA costs will be closer to R8.0M.
Although any one particular leased line costs the same whether or not it is carrying network traffic, as more and more traffic flows, so the need arises to upgrade that circuit. Hence more costs are incurred.
By way of illustration, the Uninet backbone ran at 512 Kbps until the end of July, and then was upgraded to 2,048 Kbps. Some of those new circuits sustain a load of 1,500 Kbps, and have peaked at 2,048 Kbps already.
Self-evident. Uninet pays around R200,000/yr for a maintenance agreement, and has spent a further R200,000 on recent upgrades to the firmware. The five hub routers purchased nearly a year ago cost the best part of R1M.
From the above points, it is clear that if there is no control on the way that the network is used, then there can be no control on some very significant costs.
Within 5 weeks, the growth of traffic volumes on the USA circuit was 65%. A 33% increase in capacity is expected to be installed within the next three months.
Uninet has a classic TCP/IP backbone, with enough redundant circuits such that two of the backbone circuits can fail and connectivity will be maintained.
In all except two cases, Uninet sites connect directly to a hub on the backbone.
School regional networks connect directly into a hub site, and fan out from there. It is entirely up to such a network to decide on how best to do this - few constraints are imposed by Uninet.
In all, this makes for a very resilient and reliable network.
Uninet was the first network to provide a school with a connection to the Internet. This was to St Andrew's College, Grahamstown in about 1992.
This led to a Uninet policy that permitted all Uninet sites to provide connectivity to schools. In due course, several styles of connections took place, from dial-up email-only to leased lines with full Internet facilities.
For about 4 years, the school traffic was carried free of charge. It was always touch and go as to how long this would last. By 1996 there was clearly a problem, the school traffic exceeded that of a good number of fee-paying sites, and this situation could not be allowed to continue.
Soon after Uninet introduced a traffic-based tariff in 1997, negotiations started with school networks to work out the best means of charging. What has emerged is where the school networks handle the costs of any necessary access circuit(s), and pay to Uninet the costs of their traffic. This allows a network to start up without incurring huge and committed costs, and as and when more benefits are obtained by the network, so more payment is made to Uninet.
The costs of Uninet fall into three main categories.
The sub-categories of these are
In the case of a school regional network, this might either fall away due to geographic location, or it is paid for by the school network. For Uninet sites, a cost that is averaged across all of Uninet is charged. This works out at R200/Kbps/yr.
The cost of the six backbone circuits is readily identified. In 1997, some R2.0M will be spent.
Again, an item that is simple to identify. As mentioned above, somewhere between R6.9M and R8.0M will be spent in 1997.
The main expenditure here is for upgrading and for maintenance. As mentioned above, some R400,000 will be spent on these items in 1997.
The costs of the staff of the Uninet Office are (fortunately) carried by the FRD. Ad-hoc work is sometimes paid for out of Uninet income, and some work is contracted out. This figure is climbing slowly towards R1.0M/yr.
The cost recovery model is relatively simple, given the above categories of costs. Totals for the categories of expenditure are estimated, and the income-base is used as a divisor.
Membership fees work out at about R1,000/mth. The regional school networks are exempt from such fees, and thus don't get maintenance and upgrades to their connecting router.
USA traffic is charged at R2.50/MB.
Local traffic (ie across the backbone) is charged at R0.60/MB.
There are gaps in this cost recovery. Web traffic into the Uninet caches "costs" Uninet R2.50/MB, but is "sold" at only R0.60/MB. Thus there needs to be an average hit rate of 4:1 in order to break even.
Newsfeeds are in a similar category.
The model breaks down when a connection to Uninet is run in a state of severe overload. So, Uninet has introduced a fixed-price option, which is advantageous when the load on the access circuit to Uninet exceeds an average of 40%.
The issue of cost-recovery is a large topic. Hopefully, this paper has given school networkers some insight to some of the issues, and will lead to a better understanding about the costs of running a network of the nature of Uninet. What is not yet know, however, is does such a model scale to handle a network of 25,000 schools.