Re: performance test

Johannes Eckstein wrote:
>
> I recently ran a performance test on reportlab 3.4.0 py36 and found an
> unexpected behavior:
>
> somewhere in between 500 to 600 iterations, which is equivalent to at
> most 2055 to 2465 pages of pdf output, the time spent per cycle increases:
>
> 10 in 0.67 seconds; per ch_pair:= 0,067
> 100 in 2.964 seconds; per ch_pair:= 0,029
> 200 in 7.801 seconds; per ch_pair:= 0,039
> 400 in 27.064 seconds; per ch_pair:= 0,06016
> 500 in 44.385 seconds; per ch_pair:= 0,0887
> 600 in 63.645 seconds; per ch_pair:= 0,1
> 800 in 120.183 seconds; per ch_pair:= 0,15
> 1000 in 200.063 seconds; per ch_pair:= 0,2
>
> why is the cycle time increasing, shouldn't this be linear?

I can't say I'm surprised, especially when you add a table of contents. 
Building a table of contents is essentially a quadratic process; every
change to the table of contents means you have to reprocess the rest of
components to update links and page numbers.

-- 
Tim Roberts, [email protected]
Providenza & Boekelheide, Inc.