[Cosmo-dev] Results of Hibernate Propfind initial testing

[Brian Moseley]

On 9/2/06, rletness at simdeskcorp.com <rletness at simdeskcorp.com> wrote:

> I just did a WebDAV "ls" using the cadaver client on a collection with
> 4000 items.   This essesntially is a propfind because it retrieves all the
> attributes.  Before Brian's changes on Friday it was taking a good
> 20-30 seconds, and now it takes under 10.  I confirmed that there
> are two queries issued.  One to get all the children in the collection
> and one to load all the attributes for all the children.

excellent. how did you configure hibernate to load all the attributes
for all the children?

also, is a second ls perceptibly faster? i'd hope so, since all of the
items are now loaded into the hibernate l2 cache.

> This is running against a MySQL5 db running on another box.
> So the good news is that being able to offload the query work onto
> a separate box will improve performance significantly.  I should also
> mention that you can't really do this with Cosmo JCR.  I had Cosmo
> JCR running against a  MySQL5 db and the number weren't that
> different from the Derby numbers.  The reason is that the querying and
> indexing is still done in the jvm using Lucene.

yea, we have some anecdotal evidence that suggests derby uses lots of
memory. i wonder what would happen if bobby repeated his test after
giving the jvm a gig of memory ;)

i'd also be curious to see dao layer timing comparisons between jcr
and hibernate, which would remove all of the dav layer object
instantiations.