Re: optimization/indexing
Re: optimization/indexing
- Subject: Re: optimization/indexing
- From: Jeff Schmitz <email@hidden>
- Date: Fri, 19 Dec 2008 07:55:54 -0600
A to-many relationship is what I was thinking. Maybe Team -->>
WinAnalysis. You might need to add an attribute to maintain
ordering on the destination entity (e.g. which win?) as to-many
relationships are unordered by definition (in EOF, not necessarily
reality).
I believe that's exactly what I do have, including the ordering
attribute (place).
I did make all these relationships "owns destination" relationships
since each "entry" has its very own, unshared set of them. Could that
cause problems?
Thanks!
Jeff
On Dec 18, 2008, at 10:44 PM, Chuck Hill wrote:
On Dec 18, 2008, at 8:36 PM, Jeff Schmitz wrote:
I can agree that it's more a "collection" than a "relation", but
what I don't understand is, how is a "collection" implemented
differently than a "relationship" in EOModeler? Somewhere along
the line I got the idea that using the "MutableArray" prototype was
bad juju, so if not that, and not a blob and not a relationship,
then what?
A to-many relationship is what I was thinking. Maybe Team -->>
WinAnalysis. You might need to add an attribute to maintain
ordering on the destination entity (e.g. which win?) as to-many
relationships are unordered by definition (in EOF, not necessarily
reality).
Chuck
On Dec 18, 2008, at 10:29 PM, Chuck Hill wrote:
On Dec 18, 2008, at 7:41 PM, Jeff Schmitz wrote:
By consider a different design, do you mean something like the
below (from the wiki)? Coming form an OO world, I perhaps took
the paradigm too far and chopped up my data into too many tables?
I'd consider going a step back before that and re-considering your
OO design. 65 outgoing relationships to the same object seems a
little off, even in straight Java OO. That sounds more like a
collection of objects to me.
e.g. Would denormalizing my 1 --> 65 --> 2 tables into a single
table help?
Maybe, but it might also hurt. You don't want to just blindly
assume something and start slapping your code around.
Or would a better suggestion be to use "blobs" for the 65-->2
part so in the end I'd have a relationship of 1-->1 blob? If the
blob route, can I assume you wouldn't want the optimistic locking
to check the blob for changes?
For me, that would be a last resort.
Chuck
A common experience with large and complex object model, is that
people model their objects, then do a large fetch and find out
that bringing in a large set of EO's can be really slow .
Adapt your model
When you are going to be using a relational to object mapping
library (like EOF), you should expect that this will change your
requirements; enough that you can adapt your model to fit the tool.
If fetching an EO is heavy/slow, then generally the fewer objects
you bring in, the faster your system will perform. So if you
collapse and denormalize many small tables, into a few bigger
ones, you will be doing fewer EO loads, and probably dealing less
with all that fault and relationship management of all those
little fragments and relationships; which can result in
performance savings.
You can do this a little by flattening relationships, or using
views in the database to make things appear to be flatter than
they are; or you can go right to model and actually flatten.
Arguments can be made for each, depending on your requirements.
You can even go further, and start moving complex data structures
and relationships into blobs that you manage yourself. This
offloads EOF from managing them, and often allows you to speed
things up; but the cost is more code maintenance on your part,
and of course denormalizing can negatively impact the design so
you want to be careful about how zealously you go down this path.
On Dec 17, 2008, at 12:39 PM, Chuck Hill wrote:
On Dec 17, 2008, at 9:25 AM, Jeff Schmitz wrote:
Yes, now that I think of it, there is one of these "crazy"
joins that's probably coming into play that joins each of my
7000 rows with 65 rows in a different table, so that table must
have about 450,000 rows. Any good optimization approaches for
these type of one to "very many" relationships? recursive
fetch? I can see this table getting into the many millions of
rows real fast.
I'd spend some quality time considering a different design. I
doubt it is doing a crazy join. My money would be on Mike's
prediction of insane amounts of rapidly scrolling SQL.
ERXBatchFetching can be a big help here, properly used.
Chuck
On Tuesday, December 16, 2008, at 11:09PM, "Chuck Hill" <email@hidden
> wrote:
Either some crazy joins with other tables or something you are
not
aware of is going on. 7K rows is tiny.
Chuck
On Dec 16, 2008, at 9:07 PM, Jeff Schmitz wrote:
hmm, I'm not doing an insert at all, just a read. Kind of
thought
there must be something else too though (with my limited
experience)
but figured indexing would be a good thing to do regardless
before
digging into debugging the real culprit here.
Jeff
On Dec 16, 2008, at 11:01 PM, Mike Schrag wrote:
More than a minute to insert to a 7000 row table?
Do other operations on the same DB take an appropriate
amount of
time? If not I would start looking at DNS or other
connectivity
issues. I can't fathom a FB DB sucking at that level.
this was my first thought, too ... something else is going on
here. I suspect if sql debug was turned on, you'd see tons of
faulting going on that you didn't realize and that the insert
itself is not actually the slow thing.
--
Chuck Hill Senior Consultant / VP Development
Practical WebObjects - for developers who want to increase their
overall knowledge of WebObjects or who are trying to solve specific
problems.
http://www.global-village.net/products/practical_webobjects
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Webobjects-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden