[Dev] ZODB for object persistence [Was: ZODB is not a Storage
jeremy at alum.mit.edu
Mon Nov 11 13:45:15 PST 2002
>>>>> "MS" == Morgen Sagen <morgen at osafoundation.org> writes:
MS> The 20,000/300 figures aren't related to ZODB, they're from a
MS> BerkeleyDB test I did comparing transactional/non-transactional
MS> writes (I discovered the cache was set too small so disregard
MS> those numbers). The ZODB test I wrote a while ago did this:
There are lots of tuning parameters for BerkeleyDB, and they matter a
great deal. ZODB has fewer parameters, but they probably matter just
as much. One example is the DB cache size, although this probably has
no effect on an object creation benchmark.
MS> - Create 1000 object instances (subclass of Persistence), assign
MS> properties, add to OOBTree, commit()
MS> - Repeat 10 times
MS> For ZODB3.1(B1), I got the following objects/sec:
MS> - Minimal (BerkeleyDB) = 1800/sec
MS> - Full (BerkeleyDB) = 100/sec
MS> - FileStorage = 3030/sec
These numbers all look pretty plausible to me. FileStorage is doing a
lot less work than BerkeleyDB, so I expect it will always be faster.
Note that ZODB 3.2 is going to have a significant rewrite of the
(experimental) BerkeleyDB storages. With luck, they'll become stable
rather than experimental by 3.2 final.
MS> For ZODB4, I tested only the FileStorage:
MS> - FileStorage = 1200/sec
I doubt any other storage work <0.5 wink>.
More information about the Dev