Helder Magalhães wrote:
Well, for the heap space issue, you should really consider increasing
the memory limit allocated to the JVM (the default is not even enough
for medium complexity graphics, "medium" being naturally a very
subjective measurement). ;-) Lots of threads on this subject on the
mailing list (and even a few in the bug tracker) -- guess a FAQ [1]
item for this would made sense, even if this is not related with
Batik, as it's a common Java misconception...
Well I did end up increasing the memory available to the JVM -- probably
later than I should have. That's where the "subjective" you mentioned
comes in -- I wasn't prepared to accept the amount of memory my
application would take. The fully assembled graphic to me is not really
complex (one parent graphic, and about 16 grafted-in graphics) but the
Batik-related objects (as far as I can tell from the Netbeans profiler)
are taking up about 120MB to render it -- and that's after GC.
If I were doing this all with Swing natively it would be much harder to
program to get the same feature set but it would take about 1/10th the
memory. At least the result with Batik looks great and performs well.
That's a win.
At least I know now that my program logic is correct -- and that I am
not up against some sort of Batik "bug" that I am ill-prepared to figure
out.
The good news is that all of this should be expected. :-) As you said
to be merging a document into another one, then the main document's
DOM will be updated: that explains the high number of
"DOMMutationEvent" objects. As we're at it, you may consider taking a
look at this rather interesting and powerful mechanism [2] [3]. ;-)
I'd say that the "HashMap$KeyIterator" objects correspond to pointers
to elements and element attributes of the document being merged. The
high number of both of these objects are likely coherent with the
almost 7000 nodes you said the document being merged contained
(probably is the sum of element attributes and maybe of elements
also).
This does seem to be the case. However I suspect that many of the
HashMaps floating around also have something to do with the GC process
-- a byproduct of instantiating and throwing away a lot of objects. I
just don't know enough about it to be sure.
The less good news is that I'm not sure how to workaround this. I'm
fairly convinced that you can (temporarily) disable the DOM Mutation
events from being triggered (crawling through the examples and
documentation might help) or, if you don't need them, maybe completely
disable this functionality (note that I'm also not sure if this would
break Batik internally!). Disabling the DOM Mutation events should
lead to a temporary performance improvement and memory decrease while
doing the merge. As for the "HashMap$KeyIterator", I don't think you
can do much about it, but that's also far beyond my knowledge of
Batik... ;-)
This solution did occur to me -- what would be called for is an
optimized importNode method that avoids this problem. What might also
work better is to just build an SVGDocument from scratch each time and
invoking a Builder (I don't think DOM mutations occur in that process).
From this experience what I would ask the Batik developers for is some
way of monitoring and controlling memory usage. It is easy enough to
get an object count given a Node parameter, but what is really needed is
a way to get the concrete implementation to fess up on what objects it
has instantiated and what memory they use.
When I have some time I am going to look deeper into this. For now
since I am up and running (memory is cheap, right?) I have to set this
aside and focus on my business logic.
Hope this helps,
Helder
You always seem to have a useful link or two. Thanks for the pointers.
[1] http://xmlgraphics.apache.org/batik/faq.html
[2] http://www.xml.com/pub/a/2002/10/09/mutate.html
[3]
http://www.w3.org/TR/DOM-Level-2-Events/events.html#Events-eventgroupings-mutationevents
--
Alan Deikman
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
