Monday, December 20, 2010

Slow Start and Congestion Control for a Snowbound day

I am supposed to be en route to paris via the normally excellent and affordable Eurostar train. As it happens there has been some quite awful weather including temperatues hitting -10C on the northen french bit of the route yesterday, so there were some speed restrictions and some problems with trackside kit etc etc - this is not unreasonable, except that it is probably one of the busiest weekends in the year.

I'm guessing that during a normal holiday day,
they haveabout 6 operational trains on london<->paris,
given there's about 1 train an hour and
its 2.5 hours + buffer/turnaround...

So yesterday, what we had here could be modelled as a massive TCP
congestion event (rather than packet loss, we got train loss)
with > 1/2 a windows worth of trains cancelled, plus a sudden increase in RTT.

So now today (correctly) the window is down to 1 train per RTT,
whichis 6 times less than normal, and slow start is happening,
so in about ln(6) (say 3) RTTs we'll be back up to speed -
so thats 18 hrs.

In that time, some of the latent demand will fade
(people like me probably cancel trip as we have no alernative,
or some people postpone (e.g. "cancel christmas" or "convert to Islam", since they seem to get less snow, or
perhaps find alternative...

If i was desperate, I'd use this old

Boat Train service, although the crossing might be pretty dreadful at this time of year.

There don't appear to be any flights from anywhere today, and practically none tomorrow either.

Friday, December 17, 2010

A New Internet Theory

As many of you know,
I have been striving to apply a number of more
scientific approaches to networking in general,
and the Internet specifically.

To this end, I ended up teaching the final year
undergrads this year
(who of course may not be as final as they think
given the impending part III option)
a whole bunch of underpinning theories, including
Information, Graph, Control, Queueing, and Optimisation

Now, these are all fine theories,
but they all have the twin massive disadvantages of:
1. being to hard to apply in any realistic scenario
2, being empirically falsifiable

What we need truly to model the Internet is a
simple, powerful, and incotrovertable Grand Unified Theory -
in Physics, they have such a theory and it is String.

So can we apply string theory to the Internet?

Yes, I think we can.

Firstly, obviosuly, the Internet looks a lot like string.
But (as physicists like doing) we can play cats cradle with the string.
Thus we can model the static and dyanmical evolution of the Internet
Lopology.

Closed Loops of string could be used to model packets.
Forwarding packets would be like "throwing someone a line".
Indeed, fowarding table entries, Link State Advertisement packets, AS
Path announcements, and packets themselves would all be elegantly
caputured using the same flixible little ring-like structure.

Elasticity ("how long is a piece of string" == QoS)
can finally be used to model traffic control and pricing
in a pleasingly coherent way.

String is a veryful thing in models.
Although Robust Operational Parameterized Estimators
are thicker, string is very quick to apply. And of course it
can offer a perfectly good model for multiplexing, as well as
giving straightforward explantory and predictive techniques for
understanding Kandinsky Network Operational Tantrums
which have resisted all previous analysis.

If nothing else,
when under persistent cyberattack,
the whole Internet unravels,
we can use the residue to help
wrap up christmas presents.

I shall be working on this over the break
and hope to bring convincing results
back from the brink in the new year.

Blog Archive

About Me

My photo
misery me, there is a floccipaucinihilipilification (*) of chronsynclastic infundibuli in these parts and I must therefore refer you to frank zappa instead, and go home