dropped to "just 10 Mbps when NASA used a satellite communicat
ion network," adding that "satellites orbiting at closer distances to Earth can also be adversely affected," contrary to Teledesic's claims. I am a satellite supporter, but this seems like a big hurdle to jump. Have things improved that much since last year?
Dean M. Riley
driley@hbu.edu
The main problem is with TCP's mechanisms for ensuring that the line is alive. There are timers all over TCP -- kernel timers that system administrators can modify, such as TCP_KEEPALIVE, and timers in the protocol itself. These timers do assume that the link, although possibly low-bandwidth, isn't highly low-latency.
Calculating the round-trip delay to a geosynchronous earth orbit (GEO) satellite comes out to about 0.24 second (the speed of light up to GEO orbit and back). The rest of the delay, if any, is in terrestrial gateways -- the same gateways that any satellite system must go through. So, a low earth orbit(LEO) satell
ite may be at 0.03-second latency, a GEO at 0.24 second, but both will have some overhead at the gateway.
While Teledesic's representatives like to imply that this latency problem hasn't been solved, the very small aperture terminal (VSAT) vendors such as Hughes have already modified the TCP stacks in their gateways. In essence, they "spoof" the client into thinking the connection is low-latency.
While there is yet no standard for modifying TCP to work over high-latency links, if you bought into a GEO-satellite system today, it would bring with it a solution to the latency problem. -- John Montgomery, West Coast bureau chief
Stop at Gateway, Pay Toll
