Abstract

Colin Richardson. Australia's Peak Demand for Internet Bandwidth: A modelling and forecasting methodology.

This paper focuses on theoretical and methodological issues associated with forecasting import demand for bandwidth on Australia’s principal trans-oceanic data communications link with the world Internet. "TelstraLink" now connects Paddington in Sydney with Hay in San Francisco. In 1988 it was pioneered by the University of Melbourne and NASA as a satellite circuit with a bandwidth of 64 kilobits per second. Between 1990 and 1995, it was progressively upgraded to a 4.5 megabits per second (Mbps) fibre optic circuit by the Australian Vice-Chancellors' Committee and CSIRO, which also funded the Australian Academic and Research Network (AARNet) domestic backbone.

Now jointly owned by the Telstra (Australia) and MCI (United States) Corporations, both the AARNet backbone and TelstraLink itself have been operated by Telstra Internet since July 1995. Despite massive investment to increase its bandwidth capacity more than nine-fold to 44 Mbps in just sixteen months, TelstraLink nowadays is facing excess demand pressures and Telstra Internet is receiving customer complaints during peak periods . The situation would be far worse except for a number of other Australian network providers who also have been commissioning trans-oceanic Internet links. Though individually smaller, these private links in aggregate provide Australia with at least a further 46 Mbps of international capacity.

Demand forecasting for bandwidth capacity planning at present is done as a "bottom-up" and "domestic-to-international" exercise by Australian network providers such as Telstra Internet. Typically, they add capacity at points-of-presence on their domestic Internet backbones as customer requirements become known, then sum the bandwidths and expand the overseas links pari passu. This paper lays theoretical foundations and develops methodology for a new approach, ie a "top-down" and "international-to-domestic" forecasting tool is presented to complement and enhance the present system of Internet capacity planning in Australia.

Theory drawn from the disciplines of economics and marketing suggest that Australia's consumer and business requirements for TelstraLink bandwidth at the peak hour of each month can be modelled as a single demand function. This econometric equation may have up to seven "driver variables" for which long monthly time-series of data can be readily obtained. More problematic is the dependent variable for unconstrained peak bandwidth demand, which in principle can be constructed from information on peak monthly capacity, demand, quality-of-service, and load.

The research proposes that Australia's peak demands for TelstraLink bandwidth are driven by up to seven influences, viz the Telstra Internet tariff schedule, gross domestic product (GDP), the prices of computer/modem equipment needed to access the Internet, the increasing "content-richness" of World Wide Web pages, the overall size of the world Internet (presently 13 million computers are reachable), the level of employment within Australia, and average weekly earnings. This information was used to program a forecasting tool and make (demonstration only) projections of peak monthly unconstrained import demand for TelstraLink bandwidth, with the forecasting horizon being November 1997.

The Base Case forecasting run was supplemented by a set of Perturbed Case forecasts (effectively, sensitivity analyses) based on changing the monthly growth rate test data characterising all seven driver variables in the Base Case. When more information becomes available, good estimates of the parameters of our econometric demand equation will replace the dummy own-price, income, cross-price, and other elasticities used for this set of mere demonstration forecasts.

It is suggested that a Stage 2 project be initiated. With access to appropriate demand information, this methodological exercise could form the basis of an authoritative empirical study. The benefits of deriving accurate estimates for parameters such as price, income and cross elasticities of demand include making forecasts of Australia's peak demand for TelstraLink bandwidth based on several plausible scenarios. This can only improve the practice of capacity planning and lead to more efficient outcomes.

Copyright ©1996 Colin Richardson and the La Trobe University Online Media Program