The problem of edge coverage result deviation about k-scheduler HOT 2 CLOSED

Thanks for your interest in K-Scheduler. I have been busy with interviews recently, and I will be available around mid-January.

There seems to be a misconfiguration of the K-Scheduler in your experiments. We evaluated the k-scheduler on readelf and nm before, and the results were good.

0. From your evaluation results, AFL havoc mode achieved more than 10k edges within the first hour. But K-Scheduler + AFL havoc mode only got a 6369 edge after 24 hours. I am guessing you might use two different readelf binaries. To ensure a fair and scientific comparison, please use the same binary compiled in K-Scheduler on vanilla AFL havoc mode.
1. Could you provide the Binutils source code, llvm version, wllvm version, and your final binary, along with graph info data, so I can reproduce your results and debug the configuration? I'll make sure to get back to you when I am free.
2. Could you also provide your baseline setting, including vanilla AFL, binary, and seed corpora?
3. You added additional parameters "-t 2000" to K-Scheduler. Could you explain why you added that timeout parameter and what would happen if you removed the timeout parameters?
4. You mentioned you tested 16 programs. How many programs have a similar issue (i.e., K-Scheduler significantly underperforms baseline), and how many programs have expected results (i.e., K-Scheduler can beat vanilla AFL havoc)?

Best,
Dongdong

from k-scheduler.

Resolved. I confirmed with egoterm through email that the wrong configuration (two different binaries) caused this issue.

from k-scheduler.