mario ruggier wrote:
>
> On Dec 16, 2005, at 6:33 PM, Peter Fein wrote:
>
>> mario ruggier wrote:
>>
>>> On Dec 15, 2005, at 3:17 PM, Jesus Cea wrote:
>>> Sensitivity to interruptions -- Berkely DB: very; crashes and permission
>>> problems can leave the database “wedged”, requiring journaled recovery
>>> procedures. -- FSFS: quite insensitive.
>>
>>
>> I was going to mention this.  Jesus, if you're concerned about
>> performance and are talking 1/2 TB of data, you might want to consider a
>>  RDBMS (I use PostGres) and Python middleware.  I've come to love SQL,
>> but there are a number of good ORMs - SQLObject being the foremost.
>
>
> oh yes, this reminds of the the somewhat twisted (sorry for the cheap
> pun) idea to use an sql storage backend for an object database, as they
> seem to be doing with Axiom, using sqlite:
> http://divmod.org/trac/wiki/DivmodAxiom

That's why I code my SQL and build python object from rows by hand.
It's not that hard.  Then again, my data is much wider than it is deep.
 We've got maybe a dozen tables that map (usually individually) to
objects with a few million records.

While this is a bit off topic, I think it's worth noting which types of
data sets map well to RDBMS and OODBMS models.  Your website content?
OODBMS.  Your user's mail? RDBMS.

Don't forget about the cost of object instantiation- turning rows into
python objects has a cost.  I see not needing to do this as one of
Durus/ZODBs performance advantages.  However, this can be easily dwarfed
by the wins from SQL querying.