Archive for July, 2008

Simulation and Modeling

Wednesday, July 30th, 2008

Simulation is a wonderful thing. It’s one of the foundations of mathematics.

Scale models are a good example. If you want to see what the new building will look like make a model. It isn’t going to tell you everything, the architectural models of housing projects usually look wonderful, but it’s a great example of using the external world to think for you, of externalizing your brain. Yes, you can read a description and imagine what something will look like, but making a model, reduces how much of your brain that uses, leaving more brain for other things, like deciding whether any child is really going to want to play under the gaze of a thousand anonymous windows.

The power of modeling is real. From the scale drawings my father used to make before decorating the kitchen, to the simulation that demonstrated that the proposed system for baggage handling at Denver Airport was guaranteed to fail horribly. (Unfortunately the simulation, costing a few thousand dollars, was done after the actual system was built at the cost of hundreds of millions, and then millions more in delays.) Dangerous, lengthy and expensive processes can be evaluated for a fraction of the cost of the actual experiment. Some people’s love this new power, others can’t forget what’s being lost.

What’s lost in any abstraction is the specific, the individual, everything that matters. Some people can’t get over that.

Simulations and models only work because they ignore details, knowing whether they are important details can be difficult. A scale model works fine for the forces on a house or a skyscraper, but is hopeless for a ship. Econometric models are notorious for their spurious certainty.

So the practical fitness of a model to it’s purpose is one question, but there are others. For some people it is impossible to imagine the classic absurdist math problem characters “a man”, or “a woman” without some humanity to hang on to. People differ on this.

I remember some fine junior high level course material on date arithmetic. A drawing of a set of gravestones: name, born, died. The question was: how old was each person when they died. One child faced with “James Brown: March 1823 to June 1839″ confounded his teacher by insisting on knowing why Jim died. I don’t share that need, but I rather like it. I’m glad it’s around. If we want to make more people more mathematically adept, this is the kind of fact we need to acknowledge and deal with.

Physical Data

Wednesday, July 16th, 2008

RDF stores in hardware.

Whatever else it turns out to be, the semantic web has already given us RDF. It’s here, it’s now, it rocks.

RDF says that every data structure can be modeled using three part statements consisting of Subject, Predicate and Object. Each of these being a simple piece of data that can be represented as a string.

This is a surprisingly substantive claim. It means two part statements (Subject/Property, or Name/Value) are not enough, and four part statements unnecessary. It means that in structured data RDF triples play the role that triangles play in graphics. If you can draw triangles you can draw everything. If you optimize your triangle drawing you optimize all your drawing. In networking TCP/IP packets play a similar role. Routers, switches and network cards don’t care what new applications are built on top of packets. Make passing them efficient, and everybody benefits, the cost of switching packets tends to zero.

The completeness property of triangles in graphics makes graphic cards rewarding – custom silicon that solves a specific problem insanely efficiently. RDF has similar potential. RDF can scale in hardware. The memory storing the statement can be smart enough to parse and query it. The structure is so simple, basic filtering so straightforward, that it can be done on a per triple basis in hardware, implemented directly in transistors alongside the non-volatile memory, built scaleable to plug into an RDF bus. As you add data you add processing power. Since one of the sweetest properties of RDF is the fact that it can be munged together indiscriminately – you can always add more triples without compromising what’s already there – this means we can have physical data. Physical data scales really well.