[ZODB-Dev] Storage Benchmarks

[Toby Dickenson]

On Monday 17 December 2001 4:00 pm, Barry A. Warsaw wrote:

>Because of the memory usage?
>
>Here's a thought: instead of storing the packablerevs in a Python
>dictionary, we could use a separate (temporary) Berkeley table that's
>/only/ used for packing.  We already acquire a pack-lock to prevent
>two threads from packing at the same time, so this would protect the
>pack database as well.

Sounds good.

It looks like the 'reachables' set might be a memory hog too, however  it 
looks like it should be possible to calculate that incrementally (needing 
memory proportional to the depth of the object reference graph)

>It might not be any faster, but it ought to be much more memory
>friendly.

Im happy to repeat my bechmarks on anything working in this direction.