Guest wrote:
Basically, the larger the better provided it all fits into physical memory
with no chance of paging out to disk. It's not going to fit into the L2
cache anyway....

I don't quite agree: every hash table entry imposes some administrative
overhead that offsets the benefit of having a hash table. If that overhead
gets too large, you lose.

Just where the cutoff is depends on how the hash table was implemented.
Testing can be used to detect the cut-off, but it will probably take a lot
of careful testing to discover.

So: ask the software writers. They should have an idea if there
is a sweet spot or not.

--
Anders Thulin anders*thulin.name http://www.anders.thulin.name/