- changed the load balancing algorithm to an optimal one
- ran speed test for serial/parallel with two worker processes and got much better results. With two workers for 400 bodies, the ration serial/parallel is about 1.6
- ran speed tests on David’s workstation and got up to 4.2 speedup
- compared our code to the data in camcos paper and found that:
- our code runs faster whether we fixed the order or not
- using the same reference solution, we get similar values for the “relative error vs wall clock time” when we fix the order. But when we allow our code to adapt the order, we get a better relative error for the same wall clock time (it’s about 2 times better)
