[GiNaC-list] little optimization => lots of Memory usage
axischire at gmail.com
Fri Oct 22 18:33:41 CEST 2010
i made the test, here is what i can tell you from my experiments
1) for a mapMatrix ( map<string, map<string, ex> > ) of 8524 x 8524 where
all expression where 0, the RAM usage was 6GB ~ 18% of 32GB, very close to
your stimation Alexei, you where right on that
2) however, if i repeat the experiment but let the expressions be what they
should be (which are pretty large), then RAM usage goes beyond 32GB, i
predict is something like 40GB of usage.
3) i would be complaining about the high ammount of RAM usage if it werent
because of the following peculiarity. If instead of a mtrix, i decide to
store the expresions in a system ecuation form with its corresponding
variables (in the way that all elements of a row are grouped into one big
polinomial expression), RAM usage is becomes low again!
2) and 3) have the same ammount of terms, just grouped in a different way,
it is strange, i am gonna do some more tests and any news i will post back,
Alexei, what is your opinion on my experience?
On Fri, Oct 22, 2010 at 12:34 PM, Cristobal Navarro <axischire at gmail.com>wrote:
> thanks for helping,
> i am going to do something equivalent to what you suggested, i will fill
> all expresions of the mapMatrix with zeros, and see how much RAM i consume
> because of the strings.
> ill post my results in a moment.
> best regards
> On Fri, Oct 22, 2010 at 9:53 AM, Alexei Sheplyakov <
> alexei.sheplyakov at gmail.com> wrote:
>> Hi again,
>> On Fri, Oct 22, 2010 at 12:22 AM, Cristobal Navarro <axischire at gmail.com>
>> > i kept investigating,
>> > based on the test i've made, im almost sure, 99%, that the increase in
>> > memory usage is due to the segmentation of the expressions.
>> I don't think the problem has anything to do with GiNaC. Hint: try
>> ints (instead of ex*) into that map. Check the memory usage, and compare
>> it with your calculations.
>> Use a different data structure (perhaps hash map or a sorted array), and
>> store ex instead of pointers.
>> Best regards,
>> GiNaC-list mailing list
>> GiNaC-list at ginac.de
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the GiNaC-list