Brian Candler wrote:
...
> Or you could just keep keys in RAM and use separate disk files for each
> object.

That's sounding sorta like fsdb[1]. I'm not sure I can recommend it in 
this case, though, since you pay for Marshal.dump every time you write 
an object into the database, plus locking, plus IO. I'd expect 
FSDB::Database::[]= will be orders of magnitude slower than Hash#[]= .

It will be interesting to hear any solution using mmap....

[1] http://redshift.sourceforge.net/fsdb/

-- 
       vjoel : Joel VanderWerf : path berkeley edu : 510 665 3407