Communication-Efficient Implementation of Failure Detector Classes $\diamondsuit\mathcal{Q}$ and $\diamondsuit\mathcal{P}$

Abstract

Several algorithms implementing failure detector classes $\diamondsuit\mathcal{Q}$ and $\diamondsuit\mathcal{P}$ have been proposed in the literature. The algorithm proposed by Chandra and Toueg in [2] uses a heartbeat mechanism and all-to-all communication to detect faulty processes. The algorithms proposed by Aguilera et al. in [1] and by Larrea et al. in [4] use heartbeats too, and rely on a leader-based approach. On the other hand, the algorithm proposed by Larrea et al. in [3] uses a polling —or query/reply— mechanism on a ring arrangement of processes. The leader-based and the ringbased algorithms are more e.cient than the all-to-all algorithm regarding the number of messages exchanged (linear vs. quadratic). Compared to polling, the heartbeat mechanism reduces the number of messages to the half. Therefore, a heartbeat and ring-based algorithm should outperform the former ones.

Research partially supported by the Spanish Research Council (MCyT), under grants TIC2001-1586-C03-01 and TIN2004-07474-C02-02, the Basque Government, under grant S-PE04UN19, and the Gipuzkoa Provincial Council, under grant OF-846/2004.