2 Power of Two Choices

In this lecture, we analyze the problem of scheduling n equal size tasks arriving online to n different machines. This could easily be solved with a centralized server that knows which machines are occupied and assigns all n tasks to unique machines. Instead, we want to see if it’s possible to distribute the tasks without without such coordination among the tasks. In this problem, the metric we care about is the maximum load on any given machine. We will consider two such approaches: naive random assignment, and the “power of two choice”.