Snap-Stabilization in Message-Passing Systems Brief Announcement

In this announcement, we report recent results [4] (available at http://arxiv.org/abs/0802. 1123) where we address the open problem of snap-stabilization in message-passing systems. The concept of snap-stabilization [3] offers an attractive approach to transient fault tolerance. As soon as such fault ends a snap-stabilizing protocol immediately operates correctly. Of course, not all safety predicates can be guaranteed when the system is started from an arbitrary global state. Snap-stabilization’s notion of safety is user-centric: when the user initiates a request, then the received response is correct. However, between the request and the response, the system can behave arbitrarily (except from giving an erroneous response to the user). A related well-studied concept is self-stabilization [5]. After the end of a transient fault, a selfstabilizing protocol eventually satisfies its specification. Thus, snap-stabilization offers stronger safety guarantee than self-stabilization: it may take an arbitrary long time for a self-stabilizing protocol to start behaving correctly after the fault. However, all snap-stabilizing protocols presented thus far used a high-atomicity execution model: each process is able to read the states of its neighbors and update its own state in one atomic step. It was unclear if snap-stabilization is possible in more realistic finer atomicity execution models such as message-passing systems. The contribution of this work is twofold.