Aleks,
I have been doing quite a bit of work on the memory management. In most
cases it should be a lot better but there's scope for some tuning. I'm
planning to come back to that in due course. If you want a stable
platform you should really stick with the last release rather than SVN.
With a garbage-collector there is always a trade-off between memory and
GC time. The larger the heap compared to the actual amount of heap
space used by the application the less frequently the system needs to
garbage-collect and so the faster the application. This is true at
least up to the point where the heap exceeds the available real memory
and starts to swap heavily. These days most people have several
gigabytes on their machines.
You can control the heap size by using the -H argument which takes a
heap size in mega-bytes. This is the initial size; the heap may grow
larger if needed, but using a smaller value may work better if you're
short of real memory.
Incidentally, unless you are using Poly/ML on a very obscure
architecture there is no "interpreter". It's all compiled to machine code.
Regards,
David
On 07/06/2011 16:12, Aleks Kissinger wrote:
I've just updated to the latest SVN of poly, and my compile times have
ballooned up. Prodding it a bit, it looks to me like there's a memory
leak in the interpreter.
Here's what I do to get it to happen:
1. Fire up poly and load a modest heap (59 megs) containing about 100
structures and functors. top shows poly's RAM usage goes to a little
over 59 megs.
2. Run some pretty functor-heavy code, including some code that looks
(essentially) like this:
structure Structure1 = Functor1(....);
structure Structure2 = Functor2(structure Foo = Structure1);
structure Structure3 = struct open Structure1; Structure2; end
structure Structure4 = Structure3 : SIG
Admittedly, these are some pretty big structs, and poly's RAM usage
hits over a gig, and doesn't go back down once execution is finished.
"PolyML.fullGC ();" seems to have no effect.
3. Save the heap, still< 70 megs.
4. Close and open poly w new heap, RAM usage still< 70 megs
So, the compile-time slowness seems to be poly filling the system's
RAM and starts swapping pages. I included the code above on the off
chance there's some cyclic dependency between the structs that keeps
some junk from being garbage collected.
I can give SVN links if you want to see the offending code itself, but
its part of a pretty large project, so I don't know how helpful that
would be.
Thanks!
Aleks
_______________________________________________
polyml mailing list
polyml@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml
_______________________________________________
polyml mailing list
polyml@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml
