||[Sep. 23rd, 2011|01:15 pm]
In the coffee room yesterday with michiexile, who has just started a postdoc in cloud-scale computational algebra:
michiexile: ... so we thought "that sounds like a nice simple example, let's see if we can replicate their experimental results computationally". So we tried it, and ran out of memory. So we added more memory -
pozorvlak: - and ran out of some other system resource?
[Insert ten-minute discussion of the extreme cleverness he had to employ to make the calculations tractable in the face of O(n3s) runtime and memory usage.]
pozorvlak: Impressive stuff. But could you in principle have run your existing algorithm on a bigger datacentre and got results that way?
michiexile: Maybe, but there's so little data in the example that I'd take it to cloud conferences and the big-data people would laugh at me.
pozorvlak: I'm not sure that would be a problem; there are plenty of combinatorics problems that can be stated in a couple of sentences which would absorb years of supercomputer time.
michiexile: Only years? So you only care about the wussy combinatorial problems, then?
pozorvlak: Well, "years" includes "millions of years", right?
michiexile: I suppose so...
pozorvlak: And "supercomputer" includes "galactic mass of computronium", right? Actually, that sounds like fun. We should probably use someone else's galaxy, though.
michiexile: A whole galaxy turned into computronium!?
pozorvlak: Sure. After all, this golf tournament won't schedule itself...