Covergence

Abstract

In distributed systems, there are two approaches to data consistency: consensus or convergence. The consensus approach can be implemented with algorithms such as Paxos or Raft, and it ensures strong consistency, which means making the distributed system appear as if it were not distributed and as if there were no concurrency. This approach makes the system easy to use, but it comes at the cost of performance, scalability, and the kinds of faults that can be tolerated, because every update needs to wait for a reply from other nodes before it can complete. Convergence is such a useful idea that different research communities have developed several ways of achieving it. This article looks at four variations on the theme of convergence, drawn from four areas of computer science. I have selected five fairly recent articles that provide introductions to each of the techniques for convergence.