Under the database, the main thing going on is efficient use of memory blocks and disk head movement. The idea is that the movement of values and indexing information and disk head all be aligned, in the forward direction. If you say, hey the B Tree+ is not accurate all the time, then Shannon says, how inaccurate? That determines the -iLogi total probability of some request requiring at probability i for some log(i) size, in the optimum. That is we should be able to compute the optimum break points, where a B tree expand or contract the rank of its index, getting nearly good estimates of data size. Treat the delivery of data like the optimum channel under noise. Load up a large chunks and hold them for rare requests, medium size for medium, follow the -Logi slope. Increasing larger sizes at events get rarer.
But the natural search language, semi ordered key words cleaned by by auto-spell, actually follows that quite well.
No comments:
Post a Comment