8 Things to be done

8.1 Version 1

• Currently, nbody is rather slow. The force calculation routine runs at the speed essentially the same as that of v_code , but tree constraction phase take roughly twice longer time. This is because the qsort(3) routine used in BHtree.C is less than optimal. A handcoded version would significantly improve the performance.

8.2 Version 1.1

• The source code needs some cleaning up... It's mostly not documented and not easily understandable.

• snapshot input and output should be taken from standard input and output. However, then, what shall we do with the log output? Of course, the starlab format solves the problem, but it's a bit too large...