Title:Probe Scheduling for Efficient Detection of Silent Failures

Abstract:Most discovery systems for silent failures work in two phases: a continuous monitoring phase that detects presence of failures and a localization phase that identifies the faulty element(s). This separation is important because localization requires more significant resources than detection and should be initiated only when a fault is present.
Detection, which we focus on, uses a schedule of probe packets. We propose ways to improve the system efficiency by addressing both the selection of appropriate objectives for detection time, based on costs associated with particular failures, and the efficient design of schedules, which satisfy objectives with minimum probing overhead. Our work unifies the treatment of SUM and MAX objectives and of stochastic and deterministic schedules and allows us to compare and relate these alternatives.
We define {\em memoryless schedules} -- a subclass of stochastic schedules which is simple and suitable for distributed deployment. We show that optimal memorlyess schedules can be efficiently computed by convex programs (for SUM objectives) or linear programs (for MAX objectives), and surprisingly perhaps, are guaranteed to have expected detection times that are not too far off the (NP hard) stochastic optima.
Deterministic schedules allow us to bound the maximum (rather than expected) cost of undetected faults, but like stochastic schedules, are NP hard to optimize. We propose novel efficient deterministic schedulers with analytic performance guarantees and simulate them on real networks. By relating performance to the computed memoryless optima, we can see that on many instances, our deterministic schedules are nearly optimal.