I am no expert in this subject, but the cost of doing I/O is always
going to exceed the cost of in-system memory copies, probably by
orders of magnitude. The only place where this might not be true is
using an in-memory database such as an HSQLDB mem jdbc connection.
So unless you're copying orders-of-magnitude more data than you'd be
fetching, always copy.
Remember too that a localObject only copies the object, not any of its
relationships. And the other objects are only pulled in if you
access them, which is even more fine-grained than prefetching.
So my guess is that localObject is ALWAYS orders of magnitude faster.
I suspect it'd be extremely difficult to define a situation where this
was not the case.
On 11/1/07, Marcin Skladaniec <marci..sh.com.au> wrote:
> Hi
>
> We are improving the performance of our client application (ROP). Not
> surprisingly we have found that adding prefetches did improve the
> speed significantly.
>
> I have a question though about which is about efficiency of
> localObject. We are fetching a list of objects (query with prefetches)
> to a specific one non-editable context. If the object is to be edited
> it has to be copied to another context which allows committing
> changes. After the object is copied all the related object are
> accessed, so the performance of the copying came to my mind.
>
> The test I have looks like this:
>
> CayenneContext context1;
> CayenneContext context2;
>
> SelectQuery q = new SelectQuery(Painting.class);
> q.addPrefetch(Painting.GALLERY_PROPERTY);
> q.addPrefetch(Painting. GALLERY_PROPERTY + "." + Gallery.CITY_PROPERTY);
> List l = context1.performQuery(q);
> //now the context1 contain all the records I wanted it to contain
>
> Painting p1 = (Painting) l.get(0);
> Painting p2 = (Painting) context2.localObject(p1.getObjectId(), null);
>
> //now the context2 contain only the single record I copied
> p2.getGallery().getCity();
> //now context2 contains the same objects as context1
>
> All is very quick, but my question is how it will scale when the
> relationship would be to-many and there will be hundreds related
> records.
>
> What do you think would be a threshold number of related objects which
> have to be localised over which it would be worth doing a new,
> specific select query for that single object (with prefetches). Would
> there be advantage of doing that at all ?
>
> We are using cutting edge version of cayenne (I think we use build
> about 2-3 weeks old).
>
> Marcin
>
This archive was generated by hypermail 2.0.0 : Fri Nov 02 2007 - 09:24:09 EDT