Scheduling Solutions for a Unified Approach to the Tolerance of Value and Timing Faults

Source: CiteSeer


ng currently approached. An attractive direction of pursuing tolerance to timing faults is to identify a minimal implementation of the task, whose correct execution satisfies some minimum requirements, and building upon it a fault-tolerant scheme. This "critical" part, which has to be kept as much as possible reliable and verifiable, would be called in emergency situations to provide a minimum, though degraded, acceptable level of service. The TAFT (Time Aware Fault-Tolerant) scheduling strategy [2] has been recently developed to provide a method for flexible and predictable execution of tasks with hard timing constraints (no deadline violations allowed) in presence of timing faults. It also provides a flexible implementation base to enable an easy mapping of a variety of strategies for the tolerance of faults in the value domain [3], thus configuring as a nice solution to the integrated tolerance of both timing and value faults . A TAFT component is structured in a task pair (TP), m

Download full-text


Available from: Andrea Bondavalli, Sep 18, 2013
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents an approach to include into the decision-making processes of CAUTION++ the mathematical formulation of the business models as a means to guide the reaction against overload effects. A driving decision-making criterion on resource assignment is found in the value the system obtains with the satisfaction of a service request, for a given traffic load scenario. The optimal solution for the RMU and GMU business models is the one that maximizes the total value obtained by the system. To find the optimal solution of business models, we propose a heuristic Branch and Bound algorithm that extensively searches the state of the feasible solutions by evaluating and comparing them in terms of the value obtained by the system. The space of feasible solutions can be described as the leaves of a tree, where each level corresponds to a decision that assigns a value to a decision variable for a particular type of service request. The algorithm adopts a greedy search- limiting criterion consisting in the pre-evaluation of the partial solution to be explored. If the estimated value of the optimal solution in a certain sub-tree does not seem to be enough promising, all the sub-tree is ignored during the search. This greedy decision is taken according to the best solution that the algorithm has already found, weighted by the computational cost necessary to explore the new promising solutions. This way, the algorithm is able to adaptively tune the quality of the solution with the amount of computational resources required.
    Full-text · Article · Oct 2011