David Johnson wrote:

In general we’re looking at using an RDBMS, but we’re trying to
> get a better feeling of when we can use the ZODB and when we
> should stick to the RDBMS.

We have various datasets:

   1. Transactions: 10-100 billion objects
   2. Users: 10-100 million objects
   3. Contacts: 100,000 objects
   4. Customers: 100-1000 objects

I see other good replies but one factor you must keep in mind for ZODB is that it keeps the index of all objects in memory. With 100 billion transaction objects, you're going to need a LOT of RAM and have a large startup-time.

I've always wondered why ZODB didn't offer an option of storing the object index in a BTREE, but I guess its the performance issue of potential additional disk accesses.


Zope3-users mailing list

Reply via email to