stephenrkell
Performance benchmarking
It's been absolutely ages since I did any serious performance measurement of liballocs, specifically execution time overhead.
It would be good to be able to repeat-or-beat my c.2015 SPEC CPU (2006!) numbers. :-)
That may not be trivial.
Memory performance is of interest too, of course. I've never done a thorough measurement of that. It is quite subtle of course -- what to measure? Peak RSS is one answer... probably the right thing, although of course it does not give a discount for inter-process sharing.
It may be more work, but I suggest setting up a weekly/monthly CI task to run a benchmark on the latest. There is plenty of computing time on the ER server, and it would be great to see how changes impact performance. Necessarily, this would also force contributions to keep the benchmark suite operational, which will save a lot of pain in any future paper prep
