À la Une

Soutenance de thèse Cristina Muñoz Illescas

Mme Cristina Muñoz Illescas soutiendra, en vue de l'obtention du grade de docteur ès sciences, mention informatique, sa thèse intitulée

A Distributed Event-Based System based on Fragmented-Iterated Bloom Filters


The purpose of this thesis is to design a distributed-event based system for wireless sensor networks that constructs an overlay layer using: (1) Directional Random Walks, (2) Compressed and (3) Fragmented-Iterated Bloom Filters. Moreover, we study an innovative application to use on top of our system based on Searchable Symmetric Encryption.

The dissemination of sensing data requires the participation of different sources and destinations. Typically, in an ubiquitous sensing scenario some nodes provide data and other nodes use these data as actuators. Then, a distributed event-based system may be used to exchange information. In such a system, publishers and subscribers do not have any information about each other. They depend on the event notification service to match publications with subscriptions. In distributed networks, this service is implemented using a network of broker nodes. A broker node is any node in the network that has information about any single or set of subscriptions. Publishers must contact a broker node to route events. Similarly, subscribers rely on broker nodes to save subscriptions.

The selection of broker nodes in a structured network requires the use of an overlay layer on top of the network layer. We propose to merge the network and the overlay layers of distributed event-based systems so that no other network protocol is needed. Paths between publishers, subscribers and brokers are well-defined using Directional Random Walks (DRWs). The main idea behind a DRW is that two lines in a plane cross if they are not parallel. The advantage of this strategy is that it is unnecessary to maintain the information concerning the network topology. The main consequence is that nodes, which do not actively participate in the system, do not keep any information about topology. This leads to save energy and computing resources in those devices.

With the aim of providing forwarding capabilities to broker nodes we propose to implement a novel architecture of Fragmented-Iterated Bloom Filters (FIBFs). A Bloom Filter (BF) is defined as a probabilistic data structure that efficiently manages membership of a certain number of elements. Each broker node implements FIBFs at each interface of communication. FIBFs effectively save a set of subscriptions that use conjunctive and disjunctive operations. When a publication arrives to a certain interface, the corresponding FIBFs are checked to decide if it has to be forwarded or not. A publication is forwarded if there are subscribers behind that interface, which are waiting for that specific type of events. 

Broker nodes of our distributed-event based protocol require to update the FIBFs by contacting their broker neighbors. The purpose is to decrease the number of transmitted bits to reduce the bandwidth and the latency required. Besides, this strategy also improves the performance of radios because less data is transmitted so that energy is saved. The methodology followed requires a compression algorithm. As a result of this compression, the probability of false positives of FIBFs is reduced.

Finally, we study a future application that may use our system based on Searchable Symmetric Encryption (SSE) that allows to search for documents in an encrypted domain. Secure Indexes based on BFs are used to locate the appropriate encrypted documents. We improve the construction of Indexes in terms of security and performance. Our approach relies on index blinding based on a given entropy level. Moreover, we focus on the performance evaluation of different strategies to securely populate BFs using cryptographic hash functions and a cryptographic algorithm plus a fast hash function.

Quelques images:





Date: Mardi 19 janvier 2016 à 16h00

Lieu: Battelle bâtiment A - Auditoire rez-de-chaussée

13 janvier 2016
  À la Une