gamozolabs
Use INT_MAX as persistent count
100000 as persistent count means that it is still forking every 100000 executions, that is a comparable time to fork i guess, so it is better to use INT_MAX (cannot disable fork at all for AFL++).
Btw, using this program as a bench does not really make sense, the bottleneck is the fuzzer and the pipe used to communicate the request for a new execution, which should be a negligible overhead on real programs. Maybe worth a try using an existing libfuzzer harness on a relatively small library to test, like pcre2 or similar?
He wants to measure the worst-case overhead. I think that is fine. It shows the general issue of the implementations.
However I agree a real world target like libpng or libjpegturbo added for comparison would be good (to see what the approx real-world limit of usuable cores for fuzzing on one system is)
