 |
LEOs Dance The Jitterbug
If you are invited to the broadband satellite ball, don't forget to brush up on your jitter bug and latency loop. A whole lot of satellite providers are dancing around these two important issues-which may well determine what applications businesses run on low earth orbit (LEO) and geostationary earth orbit (GEO) satellites.
While many of the latency issues associated with broadband GEO satellites are becoming better understood (see page, SIDELATENCY), broadband LEOS are still new enough to spur serious scientific debate.
GEO satellites typically maintain an orbit 22,300 miles from the earth's
surface and are tied to the earth's rotation, so that they appear to be fixed over a specific geographic area. Broadband LEO satellites will orbit only 400 to 1,000 miles from the surface and, therefore, a greater number of satellites, traveling in more complex orbital patterns, must be launched in order to cover the globe. At Teledesic that comes to a near boggling 288 birds, a downward revision from an even more ambitious startup plan with 840.
One of the great unknowns about these LEO constellations is how well they'll be able to handle variations in delay, otherwise known as variable latency or jitter. A low orbit satellite may only spend tens of seconds over a given geographical area, which means that a given transmission may be picked up and passed on by multiple satellites. And because satellite orbits are typically maintained within a range of locations, rather than precisely, the pieceparts of a single transmission can be subjected to varied delays and subsequent packet reordering.
F
red Baker, chairman of the Internet Engineering Task Force, says the "random and constantly changing distances" between satellites in a LEO constellation means that "TCP's window duration measurement algorithms will be constantly rubber-banding and snapping."
The simple way to look at it, says Baker, is that "if your application-voice, video, IP multicast service, etc.-is hurt by jitter or packet reordering, GEOS are your friend. If your TCP implementation is very robust, and jitter and reordering are not issues, then LEOS will give you lower end-to-end latency." Of course if low end-to-end latency and minimal jitter are both required, the conservative tack would be to keep the transmission on the ground. Those are the trade-offs.
But there are also those who believe LEO jitter will be no worse than that already seen terrestrially or that any problems that do arise will be readily correctable. Bary Bertiger, corparate VP and general manager of Motorola's Satellite Communications Group, says "t
here are clever ways to engineering around some of these problems," work that is still on the drawing boards. Hans-Werner Braun, a University of California research scientist and consultant to Teledesic, says his personal opinion is that LEO jitter won't vary more than "a few tens of milliseconds." That contrasts with a 202 millisecond variation he recently experienced in transmitting 500 packets from home to office (using a terrestrial route that detoured by way of New York). Of course in those systems that combine satellites and terrestrial nets, terrestrial jitter must also be taken into consideration.
Harvard consultant Scott Bradner also suggests that jitter is apt to be readily cleaned up by creating larger memory buffers in earth stations. Transmission, then, would be delayed long enough so that the playback to the user is at a constant latency. Teledesic, in fact, believes this is the way to go and says it has already incorporated the cost of this buffering in its manufacturing cost estimate
s of $1,000 per basic earth station.
This buffering, says Baker, will certainly benefit voice and videoconferencing, but it won't help the 75 percent of Internet traffic that consists of short Web and e-mail traffic. "No amount of buffering will help something that doesn't transmit enough information to load the buffer," says Baker.
Of course, there are trade-offs to everything, and even when resequencing buffers smooth out voice and video latency, they can add delay. If the jitter is modest, the accompanying delay shouldn't interfere with most interactive applications, including voice, says Craig Partridge, a BBN Technologies principal scientist researching TCP over satellites for NASA. Partridge, however, calls himself "the high priest of jitter is not a problem" and emphasizes that there are many from the voice world who lack his confidence.
Baker says while it's true buffers can help, "that's only half the story." What happens, he asks, when clouds or raindrops cause a message to
get dropped. If the average roundtrip is 300 milliseconds and the variation in latency is 200 milliseconds the user will end up waiting a second or two before retransmitting. Baker is convinced such delays will be pronounced in shorter-packet transactions, such as the Web transactions and in applications like Internet telephony.
What else can be done? Motorola's Bertiger says another way around jitter is to create what is known as an inclined orbital pattern that angles off the equator-a solution that Bertiger says is being pursued by both Celestri and Alcatel's Skybridge. The technical trade-off here is that the polar regions aren't covered. This won't require any special aiming of Motorola's computer-controlled tracking terminals. Michael St. Johns, network architect for @Home Network, says the trade-off to this type of approach is that it adds complexity to the already challenging task of airborne; and NASA officials offer their out caveat that the orbital concept has only been modeled, not actua
lly tried.
One thing to bear in mind, says St. Johns, is that even with an inclined orbit, the smaller the constellation, the greater the jitter. That's because each satellite in a smaller constellation will have to serve a larger ground footprint relative to the footprint served by a larger, denser constellation. If he's correct, Motorola,'s Celestri, with 63 LEOS, could have a greater magnitude of jitter than Teledesic with 288 birds, assuming operation at similar altitudes.
Yet another, and possibly quite serious question raised by Bradner and Baker is if TCP will become confused if the jitter for a transmission exceeds the time required to send, receive and acknowledge the transmission. Bradner says if variable latency exceeds the round-trip timing, TCP may interpret jitter as packet loss and begin a retransmittal. If this happens, large data transfer would be significantly slowed. It's one of the questions for which both experts agree more research is needed.
|
 |