I’ve done my first full day of work in Zambia now, and I’m starting to get an appreciation of the unique challenges of running an ISP in an under-developed country. The technology within the country is good (WIMAX is pretty well-developed, and there’s good Wi-Fi coverage) but the difficulty is getting data in from outside the country.
Despite the arrival of fibre connections to the outside world in recent months, satellite is still the most cost-effective means of getting ISP traffic across the border. This is mostly because satellite bandwidth can be purchased in simplex, while buying fibre bandwidth involves paying for a massively under-utilised outbound connection. Obviously the satellite connection boosts the latency, so there’s a lot of trouble doing QoS so that customers who need the latency guarantees get routed to fibre.
The fact that satellite bandwidth is competetive at all should immediately tell you that bandwidth is a significant cost to the business. The economics of selling to a market of relatively poor people don’t stack if you have to buy relatively expensive bandwidth to serve them at a ratio of anything like 1:1. Caching and mirroring would seem to help, but it isn’t 1997 any more: most of what people do on the web isn’t static. You can’t cache Facebook. Even seemingly static pages often have enough dynamic content that you can’t reliably cache them. Mirroring the likes of Google search and YouTube might help, but you can’t get far without negotiating with the big boys, and they may not have time for such a small market.
Traffic shaping is another major opportunity to save bandwidth. No matter how unpopular the idea might be with Californian Slashdotters, in these circumstances the only alternative to an ISP that attempts to prioritise email over bittorrent at peak times is an ISP that goes bust and provides no service at all.