Re: [gnugo-devel] Discussion about the new cache implementation

[Gunnar Farneback]

Inge wrote:
> What I could do is to store kilonodes instead of nodes, but who says
> that 1 million nodes would be enough for any reading?  The engine
> currently does 200.000 tactical nodes per second (figure from Gunnars
> computer).  Thus 1 million nodes is approximately 5 seconds of
> reading.  Granted, this is a lot for a single call to, say,
> owl_attack(), but not completely unthinkable.  This is especially true
> for the computers of tomorrow, where we might easily do several
> million nodes per second.

How much resolution do we need, really? It seems to me that a
logarithmic scheme would be good enough, e.g.

max(0, floor(log2(nodes/32)))

/Gunnar