André Filipe Antunes Madeira

Co-advisor: N. Santos

Efficient Implementation of Causal Consistent Transactions in the Cloud

Tese submetida para provas de mestrado em Engenharia Informática e de Computadores Instituto Superior Técnico, Universidade de Lisboa.

Abstract

The characterization of network flows is relevant for multiple applications, in particular for security appli- cations, such as the detection of covert channels in real time. Typically, the characterization of network flows is performed by registering all the packets of the relevant flows and, later, analyzing their char- acteristics, for example, performing machine learning classification based on the distribution of their sizes. However, this solution consumes many resources, affecting network performance. In this work, we evaluate the possibility of exploring the recent advances in SDN networks, programmable switches and probabilistic data structures (also called sketches) to characterize the flows in the switch itself, at line speed, thus reducing the amount of network data that need to be stored and analyzed to identify covert channels. We present a software architecture for programmable switches that allows us to char- acterize flows using two layers of filtering, each using a sketch. Our solution allows us to monitor 5K flows while keeping an accuracy of 0.95 in the detection of covert flows, representing an increase in analysis capacity of 20 times for the same amount of memory on the switch in the absence of sketches.

Publicações

Efficient Implementation of Causal Consistent Transactions in the Cloud: André Filipe Antunes Madeira; MSc Thesis. Instituto Superior Técnico, Universidade de Lisboa.; November, 2019.; Available BibTeX, MSC Thesis, and extended abstract, and mid-term report.
Identificação de Canais Encobertos no Skype usando Esboços em SDNs.: A. Madeira, D. Barradas, N. Santos and L. Rodrigues; Actas do décimo primeiro Simpósio de Informática (Inforum), Guimarães, Portugal, Sep. 2019.; Prémio melhor artigo de estudante INForum 2019.; Available BibTeX, extended report (pdf).

Luís Rodrigues