Fault-tolerant critical section management in asynchronous networks

Abstract

The topic of the paper is managing a fault-tolerant critical section in a completely asynchronous distributed network. Assume that processors may fail while using the critical section and therefore the critical section must have at least t+1 slots. (t is the maximum number of possible faulty processors).

There are several possibilities for defining a priority rule that achieves fairness among the processors. In [FLBB] processor p has a higher priority than processor q if p asked to access the critical section “before” q. In [ABDKPR] p has a higher priority than q if p used the critical section less than q. The drawback of the first rule (that motivated the second rule) is that a “fast” processor could use the critical section much more than a “slow processor”. The drawback of the second rule is that processors should use the critical section infinitely often in order to prevent deadlock. In this paper we modify the second rule to circumvent this last drawback. Processor p has higher priority than processor q only if p used the critical section less times than q, p wants to access the critical section and q “knows” this fact.

We present two algorithms which require t+1 and 2t+1 slots respectively. The second is a modification of the first which trades extra slots for simplicity and overcomes Byzantine faults as well.

Supported in part by a Weizmann fellowship, by contract ONR N00014-88-K-0166 and by a grant of Stanford's Center for Integrated Systems.

Part of this work was carried out while this author was visiting Stanford university. Supported in part by a Weizmann fellowship, by contract ONR N00014-88-K-0166 and by a grant of Stanford Center for Integrated Systems.