How to model link failures: a perception-based fault model
Ulrich Schmid · 2002
We propose a new hybrid fault model for clock synchronization and single-round (approximate) agreement in synchronous distributed systems, which accurately captures both node and link faults. Unlike conventional "global" fault models, which rest upon the total number of faulty nodes in the system, it solely relies upon the number of faults in any two non-faulty nodes' "perceptions"-conveyed by the messages from all other nodes-of the system. This way, arbitrary node and communication faults, including receiver-caused omission and time/value faults, can be modeled properly. As an example, we show that the consistent broadcast primitive (and hence the clock synchronization algorithms) of Srikanth & Toueg (1987) can be analyzed under this model. As far as link faults are concerned, our analysis reveals that as few as 4f/sub /spl Cscr/a/+2f/sub /spl Cscr/s/+2f/sub /spl Cscr/o/+1 nodes are sufficient for tolerating at most f/sub /spl Cscr/a/, f/sub /spl Cscr/s/, and f/sub /spl Cscr/o/ asymmetric, symmetric, and omission link faults at any receiving node.