[Iccrg] LT-TCP followup

[Michael Welzl]

> > 5 In Marina Del Rey (and a preceding email), Lachlan presented
> >   his theory which contradicts the intuition in item #3 above.
> >   Could it be that you *only* consider a wireless link, not
> >   the rest of the path? That could be a reason for your idea
> >   of actually increasing the rate to make sense in your setup,
> >   but not in practice
> 
> No, we consider the whole path.  The assumption is just that that the
> wireless link is the *last* hop.  I agree that this doesn't consider
> the problem of uploading from a wireless link (which is still
> work-in-progress), but it certainly covers a very large fraction of
> the current cases.

Hmm... I don't think that this makes much of a difference for what
I was trying to say; see below...


> I was deliberately trying to make it sound counter-intuitive.  If it

Sure, I get that, but my problem is: on the one hand, I don't fully
understand your proposal (I just took a look at the paper you
pointed to - the last time I read it I'm sure it was only 2 pages
long, and I failed to fully grasp it - so please forgive me, I
should probably see if I can understand the updated version).
On the other, it just doesn't intuitively make sense to me, which
I think that it should eventually also do, once I take a closer
look at it with an example.

Consider a long path with a wireless last hop. Consider a source
which sends across this path, and let's say that all of a sudden,
90% of the packets are lost due to corruption, and that situation
stays stable for a while. You argue that the sender should increase
its rate.

Now consider the first hop, which is the bottleneck, and assume that
all other flows sharing it get 100% of their packets across most of
the time (they don't traverse a wireless network).

The  part that I don't get is: why can it be acceptable for the
single sender which only gets 10% of its packets across anyway
to increase its rate, and thereby increase the chance for all others
to experience congestion?

That reminds me of the old congestion collapse scenario in one
of the early Raj Jain papers, where a source which doesn't
know about the small capacity of its last hop yet sends at
a high rate can contribute to congestion on a hop before that,
thereby causing the total throughput of the network to
decrease even when sending rates are increased.


> I'm not sure what specific part of our/John Leslie's proposal you
> don't find acceptable, but if you tell me then I'll gladly do what I
> can to fix it :)

It's not unlikely that I'll find it acceptable once I've fully understood
it! So far, I was also not aware that you're making a concrete proposal
for a certain behavior - at least back in Marina Del Rey, I thought that
you're merely trying to point out an interesting counterintutive issue
(which I didn't, and still don't fully comprehend).

In any case it would be helpful if you could explain it by means of
one or two simple examples like the one I gave above.

Cheers,
Michael