Performance vulnerability in dname compression

[vzuevsky]

We saw some automated NSEC walks were degrading performance of our HW fleet. Based on that, I had designed an exploit that exacerbated the impact. Both scenaria employ query packets with the maximal number of labels:

I profiled software performance under degradation and observed the culprit, that happens be LDNS wire compression algorithm as per this flame chart:

I have now redesigned that algorithm and would like to contribute with a patch. I have also added a doc file explaining the proposed fix.