On Wed, Jul 28, 2004 at 09:49:32PM +0100, Jon Dyte wrote:
> 1)The README in the distribution mentions using Durus for upto a
> million objects? Is that number based on some empirical testing or
> observed problem with more than that number?

That number is meant to give people a rough idea of the scalability
of Durus.  It is not based on any testing or inherent limits.

> Would you expect problems with a PersistentDict with say 450000
> key/value pairs?

You may need a fair amount of memory.  I tried creating
PersistentDict that contained 500000 Persistent instances.  It took
172 seconds to commit it and the resident size of the client went up
to about 250 MB.  The underlying filestorage is about 50 MB.
Setting _p_changed on the PersistentDict and commiting it takes
about 25 seconds and adds about 16 MB to the filestorage.

Using a PersistentDict with that many entries would be highly
suboptimal if you needed to add and remove entries frequently.

> 2) is there any intention or need at MEMS to implement BTrees like ZODB?
> Would anyone hazard a guess at how difficult/easy it would be?

Implementing BTrees should not be too difficult.  David is toying
with them right now.  He has also implemented a linked list data
structure.

  Neil