For this Kangaroo ECDLP solver, if RAM was not an issue, what is the optimal DP setting?
Would lower always be better? Small DP means you have to find more DPs.
Expected group operations remains the same no matter how you adjust the DP, right?
So what is the optimal DP setting if RAM is not an issue?
If the RAM was not a issue, it would be better to use a low DP, because high DP means long time between a collision and its detection, especially if you use many kangaroos in parallel.
But not too low, with DP = 0 ** you would have the minimum number of steps, but the generation of the the start points is much slower than the generation of the other points of the path. Let's say that the cost of generating a start point is about x50 the cost of generating the next point with a single jump, with an average length of 10k points (about 2^13) you should have a good value. Then DP = 12 / 13 / 14, not more.
** A note: if you use equivalence classes with DP = 0, you need only sqrt(2).sqrt(N) steps, this is the expected group operations.