Cas Cs 537 :

Distributed computing often involves a set of n processes arriving at a consensus/agreement on some bit v, in a de-centralized fashion. The problem is particularly interesting in the presence of Byzantine failures, i.e., an adversary may adaptively corrupt upto t among these n processes and cause them to behave arbitrarily in the consensus protocol. To elaborate, these processes may either completely avoid participation or send differing values to different correct processes. In short, these faulty processes aim to delay the achievement of consensus in the system as far as possible. In this lecture we shall make a few limiting assumptions on these faulty processes. These shall be introduced as and when required. In this lecture we introduce randomized Byzantine consensus protocols, where the processes are completely asynchronous. This was shown to be impossible in the deterministic setting. Note that a trivial byzantine agreement protocol simply agrees to 0 irrespective of its current state. To avoid such trivial protocols, we require two properties of any randomized byzantine agreement protocol (same as in [3]):