>Craig Partidge said that many satellite people say
>"we agree" with the goal of error-free links, a few
>do not.
Although the context in which this attribution was
provided doesn't really support my doing so, I'm hoping
it was offered in the spirit of:
"Enough already, let's stop talking about lossy
links and move on!" :o)
The wording and what it implies, however, really bug
me regardless of the intent.
Speaking as one of the folks undoubtably categorized
as amongst the "few", I'm sure the above sounded
good at the time, but it simply isn't true.
Unless including the existence of transmission errors
as part of the characterization (some satellite
environments have them, some don't) needs to become
a consensus item, I'd like to offer a one-time only,
alternative view.
First, does anyone honestly believe that some
people/communities are actually *opposed* to the GOAL
of providing a error-free links whenever possible?
If so, what could possibly be the motivation?
A goal is an ideal - something to strive for,
something you have yet to realize and for which there
is no guarantee of achievement.
I offer the following as (from my perspective) a more
accurate statement:
"while all satellite people say 'we agree' with the
goal of error-free links, for some it is not yet
always possible (practical ?) to realize that ideal"
Change the satellite to terrestrial RF/Wireless and
I believe the same applies.
The original statement leads one to believe that
lossy links are merely a lifestyle choice, and that
the only reason that they haven't been eradicated is
because they are preferred by small roving bands of
Ludites.
If one choose not to deal with lossy links, then they
don't have to; If one is able to eliminate the problem
by cleaning up the link at any cost (power, latency,
effective bandwidth are not a concern), that's cool too.
But please don't assume or assert that if there are still
systems that do not achieve error-free links that it is
because their wasn't sufficient desire.
Realistically, an error-free link *is* the cost of entry
if you want to maximize the possible performance of a TCP
connection - reduction in transmission rate due to legitimate
network congestion is a desirable trait, reductions to false
congestion events is an undesirable trait; TCP works in a
mixed-loss environment, just not terribly well (due to false
congestion responses). When you add a long-delay path, things
are even worse.
Best practice (common or otherwise) is to provide as clean
a link as possible; In some cases the error rate will still
be substantial for TCP ( >> 10^-8); For each decrease in
error rate you achieve, you pay in power, effective bandwidth
and/or latency.
Getting link-layer mechanisms to *transparently* integrate
with transport mechanisms is not always trivial. In some
inherently noisy environments FEC and link-ARQ may impose
somewhat arbitrary (unrelated to the rest of the network)
delays to the transport protocol; This might actually be
worse than trying to living with a link that is slightly
less than perfect. Security measures may prevent doing
more some clever, non-transparent transport-centric things
at the link layer that violate layering.
Also, unfortunate, unexpected even and sometimes even
tragic things do happen before or after a satellite goes
operational. These events aren't always quite catastrophic
enough to render the satellite useless, but can noticeably
reduce its communications capabilities. Still being able
to transfer data from on-board instruments/experiments in
such a situation is generally considered to be a good
thing. The same applies with a communications satellite.
Nobody actually expects TCP to *ever* treat any losses
as anything other than congestion - it probably wouldn't
even merit consideration as a goal (using the definition
I provide above). It *is* difficult and impacts what is
a currently a small portion of the Internet community;
Those are perhaps good enough reasons for opting not to
deal with the issue directly. It's just not a popular
topic.
However transmission errors, like high-latencies and
bandwidth asymmetry ARE characteristic of *some*
satellite environments, and mixed loss environments
provide interesting, relevant areas for research for
those so inclined. Nothing related is going to be
proposed or advocated as part of this WG - it isn't
within the charter. If you don't have lossy links
as a characteristic of your systems, terrific. Some
other folks don't have long-delays or bandwidth
asymmetry; The environment and the industry can not
be made to fit into monolithic characterizations.
For the portion of the satellite community (the pure
SATCOM folks) competing with fiber-optics for market
and mind-share, error-free links are probably *the*
most important design point for new systems - at the
very least for marketing purposes;
But that being the case, why stop at marginalization
of lossy links?
Since, if required cost, complexity, bandwidth or power
are not an issue, one can engineer out virtually all
characteristics of a satellite system that distinguishes
it from a wired terrestrial system, would anyone object
to the following generalization:
"many satellite people say 'we agree' with the
goal of low-latency, symmetric paths, a few do not."
Once we needlessly begin segregating portions of the
community, even if it is done with the best of intentions,
where does it stop?
My height of my soapbox is giving me nosebleeds.
Sorry - but I would have ended up saying it in Munich, so
I figured it was appropriate to present an alternative
viewpoint here. :o)
Eric
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:30 EST