Team sardes

Overall Objectives
Scientific Foundations
New Results
Contracts and Grants with Industry

Section: New Results

System support: Protocols for resilient systems

Participants : Vivien Quéma, Alessio Pace.

The goal of this work is to develop efficient protocols for building fault resilient systems. We have targeted this year two main aspects: the development of a new abstraction for constructing efficient state machine replication protocols that are tolerant to Byzantine failurs (BFT protocols), and the development of a new gossip protocol that is able to limit the dissemination of spam messages.

Modern BFT protocol implementations are complex pieces of software, encompassing sychronization, networking and crytography, which are notirously difficult to develop, test and prove correct. In collaboration with R. Guerraoui's team at EPFL, we have developed a new abstraction to simplify the development of efficient and correct BFT protocols, where a BFT protocol can be seen as a composition of instances of our abstraction. We have shown the benefits of our abstraction by developing two new BFT protocols: Azyzzyva and Aliph. Azyzzyva is a protocol that mimics Zyzzyva (a well-known BFT protocol developed at the U. of Texas at Austin) in best case situations, and that uses PBFT (another well-know BFT protocol developed at MIT) to cover worst case situations. Aliph is a new BFT protocol that outperforms previous BFT protocols both in terms of latency and throughput. This work has been presented at the Eurosys 2010 conference, where it won the Best Paper Award [35] .

Gossip protocols are an efficient and reliable way to disseminate information. These protocols have nevertheless a drawback: they are unable to limit the dissemination of spam messages. Indeed, messages are redundantly disseminated in the network and it is enough that a small subset of nodes forward spam messages to have them received by a majority of nodes. In collaboration with Sonia Ben Mkhtar from the LIRIS laboratory in Lyon, we have developed FireSpam, a gossiping protocol that is able to limit spam dissemination. In order to make it usable in practice, we designed FireSpam in the BAR model. This model takes into account selfish and malicious behaviors. We have shown in simulations that FireSpam drastically limits the dissemination of spam messages, while still ensuring reliable dissemination of good messages.


Logo Inria