Ver Post

Apresentação publica de proposta de Tese - CAT

28 Dezembro 2015, 10:30 - Maria João Silva Carvalho

Candidato: Daniel Silvestre
Título: Fault-tolerant Stochastic Distributed Systems

Orientação:
    Professor Carlos Silvestre, DEEC/IST
    Professor João Hespanha, UCSB

Membros da CAT:
    Professor Paulo Oliveira, DEM/IST.
    Professor João Xavier, DEEC/IST.
    Professor João Hespanha, UCSB
    Professor Carlos Silvestre, DEEC/IST

Data: 30/12/2015 às 18:00, Sala de Reuniões do DEEC
Abstract:

The problem of designing fault-tolerant distributed systems is presented in this preliminary report of the doctoral thesis. In particular, focus is given to addressing the case where the actions of the nodes or their interaction is stochastic. The main objective is to detect and identify faults and upgrade current systems so that they are resilient to crash-type faults and the presence of malicious nodes in pursuit of exploiting the network. The analysis is divided into three components, namely, addressing crash-type faults, malicious agents and computational solutions to performing fault detection. Crash-type faults, where the affected component seize to perform its task, are tackled in this report by introducing stochastic decisions in the deterministic distributed algorithms. Prime importance is placed on providing guarantees and rates of convergence for the steady-state solution. The explored scenario of applicability is related to distributed systems in the sense that a social network is analyzed and given its stochastic counterpart. The algorithm is capable of dealing with packet drops, delays, medium access competition, and, in particular, nodes failing and/or leaving the network. The concept of Set-Valued Observers (SVOs) is used as a tool to detect faults in a worstcase scenario, i.e., one where a malicious agent can select the most unfavorable sequence of communications and inject a signal of arbitrary magnitude. For other types of faults, it is introduced the concept of Stochastic Set-Valued Observers (SSVOs) which produce an a-confidence set where the state is known to belong with at least probability 1- a. By exploiting the structure of the consensus algorithms, results regarding computational complexity are presented by reducing the number of allowed interactions in the model. The main drawback of using SVOs for fault detection is their computational burden. By resorting to left-coprime factorizations, it is shown in this report how one can reduce the complexity by designing a dead-beat gain for the observer. In addition, changing the factorization can deal with detectable systems (i.e., system with unobservable but stable eigenvalues). Such a result plays a key role in the domain of Cyber-Physical Systems (CPSs), which otherwise would have SVOs producing divergent estimates.