>hoping for sunny days. Craig Partridge notes that "we went to a variety of
>industry meetings and talked about handling error rate issues and were
>told, point blank, by industry reps that the satellite community thought
>that was a waste of time because their (near-term) goal was to make
>satellite error rates consistent with fiber". It is my suspicion that these
>folk are either marketeers or incurable optimists.
Being a satcom industry insider I would tend to agree with that comment.
The methods used to overcome satellite channel errors will impact
protocols such as TCP.
>
>The fact of life is that space links *are not* fiber links -- "when they're
>good they're very very good, but when they're bad they're horrid". Many of
>us feel that a more robust solution would be a local approach which
>exploits the fact that we generally have very clear indicators when the
>space link is going south. Providing some inter-layer signalling whereby
>the space link can tell TCP "hey, that loss wasn't congestion, it was a
>fade" may not be a bad idea at all. With such an approach, the TCP
>retransmission control becomes an extension of the FEC, not an independent
>entity.
I would also think that, based on the types of coding and access schemes
being proposed for the Ka-Band systems, the problem is even more
complex than that. There have been a number of proponents of
schemes where the amount of coding applied is varied depending
on the perceived link margin.
This leads to interesting complications when you are in fact congested
(due to a narrower channel when coding is increased but carrier rate
is not) as well as have losses due to errors. This is in addition to
interesting
problems with varying assigned bandwidths due to a mutliple access
channel. In particular this will be seen with return channels on the
asymetrical systems such as Astra's Ka return channel service.
I guess my point is, that if you are going to fix TCP for satellite, you
might as well
fix it very well - because by the time it gets widely distributed, most of
the
people using TCP over satellite, will be using services and technologies
that are not available today.
(my $0.02)
Regards
Ian McEachern
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:37 EST