Serge Knystautas wrote:
This is turning into a bigger refactor than expected, so wanted to run
ideas past everyone...
Out of idle curiosity, why does Velocity maintain a pool of 20 parsers?
Is instantiating a new Parser object via 'new' actually that costly?
FreeMarker just instantiates them via something like:
FMParser parser = new FMParser(theReader, <and some other params>);
There is no pooling. If the pooling doesn't actually buy anything
significant, then maybe what you really want to do is just get rid of it
completely.
(Thought that comment could be helpful....)
Jonathan Revusky
--
lead developer, FreeMarker project, http://freemarker.org/
a) I realize that someone might have used SimplePool in their app, so
renaming the impl and making SimplePool an interface would create an
upgrade problem. so creating new classes.
b) the class is SimplePool and the logic and field is called
parserPool. I decided the new classes should use parser pool in the
names.
c) pools usually operate with a factory concept, which is hard with
how the current code is structured (the pool owner populates and
instantiates extra instances if there's a problem). I have to
simulate the factory concept with a callback interface, or something
like that, which is odd to pass since velocity you usually just
initialize with the runtimeservices.
d) code formatting... I don't suppose anyone has the Eclipse code
formatting rules? I'm trying to follow conventions as much as
possible, but if someone can share that, that would be very helpful.
e) I'll have the commons pool impl available as well, though I doubt
that would get added into the engine branch. Where do you think we
should stick that?
--
Serge Knystautas
Lokitech >> software . strategy . design >> http://www.lokitech.com
p. 301.656.5501
e. [EMAIL PROTECTED]
On 3/2/06, Will Glass-Husain <[EMAIL PROTECTED]> wrote:
absolutely. I think for most cases the current implementation is sufficient,
but for your high-rate of create/destroy of runtime instances I can see why
this might be a problem.
WILL
On 3/1/06, Serge Knystautas <[EMAIL PROTECTED]> wrote:
I did some memory profiling, and for each runtime instance, by default
it creates 20 parsers. According to yourkit, each parser uses just
over 100kb heap space, so each runtime instance is costing me 2mb heap
space.
I can simply reduce the # of parsers, but the problem is that simple
pool is a very, very primitive pool implementation... it holds a
constant number parsers and if you exceed, it simply creates and
disposes a parser (ballpark of 1ms per parser instantiation in my
benchmarks).
I'd like to convert simplepool to an interface and then make the
current behavior an impl so that you could swap in something like
commons pool. Is there interest in such a patch?
--
Serge Knystautas
Lokitech >> software . strategy . design >> http://www.lokitech.com
p. 301.656.5501
e. [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
--
_______________________________________
Forio Business Simulations
Will Glass-Husain
phone (415) 440-7500 x89
mobile (415) 235-4293
[EMAIL PROTECTED]
www.forio.com
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
