Stochastic bisection
by Jan Kara at SUSE Labs Conference 2022
Thanks to our conference sponsors, ARM and HPE, and our hosting sponsor and BigBlueButton provider, Blindside Networks.
During Hackweek in December 2021 I have implemented support for stochastic bisection in git. That is a bisection which accommodates the fact that during bisection the decisions about commit being good / bad are not 100% correct. For example if the failure is difficult to reproduce or if you are doing a performance testing and results tend to be noisy. In this session I want to present the model I have picked to mathematically represent the situation, what are the inputs / outputs of the stochastic bisection algorithm, and discuss how it fits people's needs or which improvements could be done to make it more useful.