Adiar 2.0.0-beta.3 : Experiment Data

In our experiments for Adiar v2.0-beta.3 (which introduces the levelized randim access optimization), we have measured running time with different thresholds for using the new optimization:

pq : The new algorithm is turned off.

auto : The new algorithm is used if the width of the narrowest input (that is canonical) is smaller than 2 MiB.

ra : The new algorithm is always used.

These measurements were (like in previous measurements) done on the Grendel cluster on CSCAA (https://www.cscaa.dk/). These measurements were done two times - once in Summer 2023 and repeated (with similar results) in Fall the same year. The repetition of these experiments were to exclude the possibility the machines were in a bad state during the summer.

time.ods is a Libre Office Calc document with all of the data analysis.

The raw data is available in the following .zip files

time_summer23.zip and time_fall23.zip contain the measurements of the running time.

stats.zip contains separate runs where additional statistics are gathered (whose running time is not of interest and even potentially misleading)

To explain an odd slowdown in the above, we timed the Apply and the Reduce algorithms separately.

split.ods is a Libre Office Calc document with all of the data analysis from this second experiment.

split.zip contains the raw measurements from this second experiment.

The code for the benchmarks can be found at https://github.com/SSoelvsten/bdd-benchmark.