[r-t] Big search

[Don Morrison]

Mark Davies wrote:
> that's just over 100,000 per second throughout the search.

Something's not right. When I run something long enough that it gets up a good 
head of steam and by amortization the startup costs (JIT compiler mostly, I 
think) become negligible, which happens in under a minute, I'm typically seeing 
node rates in the 0.5 to 5 million per second range. But surely you are running 
much faster: you're hand-coded, and paying a lot of attention to speed, and 
probably running on a faster processor. I'm using a JIT compiled language, and, 
while trying not to be profligate of resources, am not too worried about doing a 
modest amount of computation for every node and just using easy to deal with 
data structures.

Making the comparison even more surprising, I typically get the best node rates 
when there's little falseness, as with Bristol. I have to visit many more nodes, 
of course, but because there's so little book keeping to do on each visitation 
they tend to be particularly fast. I presume the same thing holds for you.

I wonder if we're counting radically different things when we say "visiting a 
node"? Or perhaps we're accounting for IO differently? Or maybe I've just got a 
bug, or am doing my arithmetic wrong?





-- 
Don Morrison <dfm at ringing.org>
"This habit of ours of seeing everything in the mind through the metaphor
of vision (a habit we succumbed to twice in this very sentence) is a major
source of distortion and confusion, as we shall see."
                               -- Daniel Dennett, _Consciousness Explained_