A problem for "Web as distributed programming paradigm"

Eugene Leitl Eugene.Leitl@lrz.uni-muenchen.de
Fri, 24 Aug 2001 19:33:38 +0200 (MET DST)

On Fri, 24 Aug 2001, Clay Shirky wrote:

> In fact (WARNING: THIS POST IS LEAVING EARTH), I'll argue that IPC is
> more than a hack, its a fact of life, woven into the fabric of the
> universe.

I thought the universe would favour largely locally coupled concurrent
message passing.

> One of the really interesting questions in biology is why cells are the
> size they are. An elephant has order 10^15 cells, a nematode worm has
> order 10^2 cells, but the cells themselves are roughly the same

10^3 for C. elegans.

> size. At a certain point, it was evolutionarily more efficient to grow

People doing patch clamping on giant squid axons or C. elegans would beg
to disagree. Some bacteria and algae cells are large, in the human sense
of the world.

> by having multiple cells than having the same number of cells grow
> larger.
> Now one of the ramifications of this is that cells need some way to
> pass information between one another, if they are to act in concert,

Notice that that information message passing is purely local, strongly
reminding one of the cellular automaton paradigm, cells aligned on a noisy
3d grid. All long range interactions arise from iterated interaction of
the local overlapping light cones. This is also true for most physical
systems, of which biology is just an instance.

> and a common theme in this information passing is that there is a
> membrane, and inside the membrane there is a really rich chemical
> environment with all sorts of processes going on all the time, but
> between any two cells there is a really limited set of information
> channels. To take one obvious example, the huge preponderance of the
> chemical work that goes on in the brain goes on inside the neurons
> themselves, while the information that passes between neurons is all
> done with just a few ions, Na+, K+, and Ca++.

That's a severely simplified picture. The brain is riddled with receptors
for funky molecules (hundreds if not thousands of them), and you see the
CNS receptors gobbling up large numbers of genome with a variety of ion
channels. Lots of molecular machinery encoded there. Ions are actually
mostly used for encoding switchable state, because you can pump them
easily, and the gradient across membrane riddled with switchable ion
channels can collapse relatively quickly.

> So, although the majority of chemical work is going on inside the
> neuron membranes, the complexity of the brain itself is a product of
> the system as a whole, both the local neuron state and the 'mere' ion
> (IPC) channels.  Whatever it is those neurons are doing, they are not
> thinking -- it takes neurons plus synapses (local processing plus IPC)
> to think.

Oversimplification alert. However tempting, explaining the world in terms
of nails has not been found agreeable to experience.

> So we have a system where membrane-bound units with simple IPC create
> complexity at a level _above_ the unit, and this theme is repeated up
> and down the scale. Within cells there are organelles -- mitochondria,

Conservation of units and clustering them in hierarchical assembly is one
of the most often recurring themes of biology.

> nucleus, Golgi apparatus -- that are themselves membrane-bounded with
> reduced channels of information flow, and the membranes of these cells
> have two cells walls, between which the chemical reactions are
> different yet again.

Membranes are compartments, where you insulate aspects of the system which
would otherwise interefere, either by chemical coupling, or by diffusion.

> On the way up, organs are membrane-bounded as well, and communicate
> through a reduced set of chemical signals, and organisms are
> membrane-bound. Most of the biological complexity in an ant colony
> goes on inside the ants, while inter-ant communication goes on with
> only a handful of pheremones and a bit of antennae rubbing. And yet

So what, it's a tight code.

> the behavior of the ant colony cannot be explained by the complexity
> of the ants alone. As poor an IPC channel as pheromones are, it is the
> individual ants plus ant IPC that make the ant colony.

Human designs do very poorly as far as emergent control is concerned.

> Even outside the biological world, systems recapitulate this
> membrane+restricted channel all the time. Alphabets and money are two
> of the best IPC channels we've ever thought of. Now I'm not claiming
> that REST is the right instantiation of the Right Answer, but I am
> willing to say that not only is IPC not going to be displaced soon by
> a distributed computing paradigm, IPC is not going to displaced ever,
> because it is a true thing about the universe, not just a placeholder
> hack.

IPC is message passing, right? If yes, I agree, that's why PVM is ailing
and MPI is dominating the scene.

-- Eugen* Leitl <a href="http://www.lrz.de/~ui22204/">leitl</a>
ICBMTO  : N48 10'07'' E011 33'53'' http://www.lrz.de/~ui22204
57F9CFD3: ED90 0433 EB74 E4A9 537F CFF5 86E7 629B 57F9 CFD3