Files

Abstract

Protocols which solve agreement problems are essential building blocks for fault tolerant distributed applications. While many protocols have been published, little has been done to analyze their performance. This paper represents a starting point for such studies, by focusing on the consensus problem, a problem related to most other agreement problems. The paper compares the latency of two consensus algorithms designed for the asynchronous model with failure detectors: the Paxos algorithm and the Chandra-Toueg algorithm. We varied the number of processes which take part in the execution. Moreover, we evaluated the latency in different classes of runs: (1) runs with no failures nor failure suspicions, (2) runs with failures but no wrong suspicions. We determined the latency by measurements on a cluster of PCs interconnected with a 100 Mbps Ethernet network. We found that the Paxos algorithm is more efficient than the Chandra-Toueg algorithm when the process that coordinates the first round of the protocol crashes. The two algorithms have almost the same performance in all other cases.

Details

PDF