Packet switching and nanotransistors

by Allen B. Downey

In recognition of the finite lifespan of the Sun, I will only take on two of the questions Veera posed.

1. Packet-switching, what's the big stinking deal?

It's all about sharing. The alternative to packet-switching is circuit-switching, which works like the phone network. When you dial a number, the phone network creates a circuit between you and the destination, and that circuit is dedicated to you and only you for the duration of the call. (At least, this is how the phone network _used_ to work.)

There are some nice things about circuit-switching, but the biggest drawback is that it is not efficient. Telephone conversations actually take up a very small part of the circuit's capacity, unless my sister is one of the participants. In other cases, the extra capacity is wasted.

That means that you pay for capacity you're not using, and the phone company has to provide a lot of extra resources (links and switches).

The genius of packet-switching is that it allows many users to share existing links. Continuing the telephone example, a packet-switching phone network breaks conversations into packets that are about 1/100th of a second long.

In digital form, these packets can be transmitted on the network in much less than 1/100th of a second. In fact, on current technology (an OC-12 link) they takes a few millionths of a second. By interleaving packets, a single link can carry tens of thousands of conversations simultaneously (and that's a conservative estimate).

There is, however, one dread harbinger of doom hanging over this otherwise utopic vision. Queueing! Or, in some dictionaries, Queuing! Or as we call it here in the States, Lining Up!

When you share, sometimes you have to wait. Some waiting is ok ---delays shorter than 1/10th of a second are (mostly) imperceptible to our slovenly human ears. But if many packets are delayed longer than that, or any are lost altogether, the perceived quality of the connection starts to suffer.

Contrary to Veera's vaguely derogatory comments about Queueing Theory, it turns out that queueing in packet-switched networks is a hugely important topic that attracts the brightest and best researchers in computer science, each of whom is held in awe by the computer science community and the world at large. It is only a matter of time before the people in Stockholm come to their senses and create the Nobel Prize for Queueing.

2) Are nanotransistors the end of Moore's law?

I will answer this by invoking Moore's meta-law, which says that every time there is a breakthrough in transistor technology, someone proclaims the end of Moore's law.

Breakthroughs don't violate Moore's law; Moore's law is _about_ breakthroughs. Moore's observation was that over a long enough time scale a sequence of breakthroughs (which occur at unpredictable intervals) can be described by a smooth function. Fortunately for those of us that like to crank gigaflops, that functions turns out to be exponential. I mean, granted, it's no Ackerman function, but so far it's treating us pretty well.

Cheers,
Allen