[Shootout-list] Two notes: Pascal and CMUCL
Bengt Kleberg
bengt.kleberg@ericsson.com
Fri, 22 Oct 2004 18:54:55 +0200
Isaac Gouy wrote:
...deleted
>>for instance, with a wider range of n we could get
>>more information about the tested languages.
>>yes, this do require that the tests are can give such added
>>information. i try to design such tests.
>
>
> This has something to do with the design of tests but not much.
i disagree.
reason: if i design a test so that oo langugage objects are created a
few at a time, and destroyed directly afterwards, it will not be
possible to measure object creation times in a low memory situation.
(unless there is something wrong with the way the objects are destroyed.
but then i am designing for languges that are defect, not what i want.
that beeing properly memory reclaimed oo langugage objects behaviour
with low memory.)
> It has a lot to do with the very practical matter of how long it takes
> to run the tests.
i agree.
perhaps it is possible to discuss how much time it is possible to spend
upon a test, before we decide that it will take too much time to measure
more n's?
> Apparently the charts on http://www.lissett.com/ben/ are based on 30
> values of N (20 in the first fifth of the chart).
it seems to me that the charts need these 30 measurements to get a
reliable value for the steady state (what i call medium n). the shootout
has 4 n's to get the same steady state value. it appears that not all 30
values are needed.
it would be interesting to discuss what the shootout need in terms of
small n's, medium n's, and large n's. this is best done after we know
how much time we have. imho.
bengt