From: Beman Dawes (bdawes_at_[hidden])
Date: 2001-04-18 11:58:16
At 08:01 AM 4/18/2001, Geurt Vos wrote:
>> I wouldn't call a total of 10 elements as representative...
>I know :)
>Anyways, I created a (I hope) more appropriate test. This time
>integers only. Pushed onto the queue are 10,000 and 100,000
>random integer values. Then 5 equivalent values are pushed,
>and the time this takes is measured. It was tested for multiset,
>boost::priority_queue and fibonacci_heap.
Since memory caches affect timings so much, it is often a good idea to also
test with even larger n and also with objects larger than ints. Not that
ints (or pointers) aren't interesting, but larger objects are interesting
Bjarne Stroustrup once mentioned that he distrusts timings that don't show
a knee in a size vs time graph, because lack of slowdown at larger sizes
often means there weren't enough test cases to get out of the cache.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk