Congratulations to Paolo Romano, awarded in the category of Researcher.
Congratulations to André Filipe Pessoa Negrão for successfully defending his PhD thesis:
Interest Aware Consistency Management and Cloud Aware Resource Provisioning for Distributed Interactive Applications
Distributed Interactive Applications (DIAs) enable geographically distributed users to interact in real time over the Internet through a shared application state. Due to their interactive nature, DIAs require the continuous dissemination and processing of potentially large amounts of data that must reach users timely and consistently. This faces application designers with two main challenges. First, the geographically distributed users are connected through networks with limited bandwidth. For that reason, it is not feasible to propagate every update to every user. Intelligent mechanisms are, thus, required to make sure that users receive the information that actually matters to them. Second, in large scale DIAs such as MMOGs, the number of concurrent users is highly dynamic, making it difficult to predict the exact number of resources necessary to efficiently provision the application. As a result, application operators tend to adopt pessimistic measures by deploying static infrastructures in which the number of resources is based on worst case predictions of load. The result is an over-provisioned computing environment in which some resources are idle for long periods of the time, leading to unnecessary operational expenses.
In this thesis, we address these issues by proposing a framework built upon two core elements: i) interest aware consistency management and ii) cloud aware resource provisioning. Our framework employs a network efficient and flexible consistency model that propagates information to users according to their interest in the different objects of the shared application state. Updates considered relevant to the user’s current task are propagated promptly; less relevant updates are postponed for a configurable time interval. Postponed updates are, then, subject to optimization strategies to improve the efficiency of update propagation. Additionally, our framework provides a dynamic resource management infrastructure for DIAs. The infrastructure makes use of a hybrid resource pool comprising privately owned resources as well as public resources acquired from a public cloud. Servers are acquired from the resource pool only when necessary and removed when they are no longer required. Within this infrastructure, we employ hybrid and task based load distribution mechanisms to improve the cost-effectiveness of the system.
Congratulations to Leila Sharifi for successfully defending her PhD thesis:
Energy-aware Service Provisioning in P2P-assisted Cloud Ecosystems
Energy has emerged as a ﬁrst-class computing resource in modern systems. The trend has primarily led to the strong focus on reducing the energy consumption of data centers, coupled with the growing awareness of the adverse impact on the environment due to data centers. This has led to a strong focus on energy management for server class systems.
In this work, we intend to address the energy-aware service provisioning in P2P-assisted cloud ecosystems, leveraging economics-inspired mechanisms. Toward this goal, we addressed a number of challenges. To frame an energy aware service provisioning mechanism in the P2P-assisted cloud, ﬁrst, we need to compare the energy consumption of each individual service in P2P-cloud and data centers. However, in the procedure of decreasing the energy consumption of cloud services, we may be trapped with the performance violation. Therefore, we need to formulate a performance aware energy analysis metric, conceptualized across the service provisioning stack. We leverage this metric to derive energy analysis framework.
Then, we sketch a framework to analyze the energy eﬀectiveness in P2P-cloud and data center platforms to choose the right service platform, according to the performance and energy characteristics. This framework maps energy from the hardware oblivious, top level to the particular hardware setting in the bottom layer of the stack.
Afterward, we introduce an economics-inspired mechanism to increase the energy eﬀectiveness in the P2P-assisted cloud platform as well as moving toward a greener ICT for ICT for a greener ecosystem.
Congratulations to Xavier Araújo Morgado Vilaça for successfully defending his PhD thesis:
Sustaining Cooperation in Dependable Systems: a Game Theorectical Approach
A dependable distributed system is composed of different processes that execute a distributed protocol to provide some reliable distributed service. Typically, one assumes that all processes cooperate by executing the specified protocol, unless faults occur; if the processes do not cooperate, then the service that the system is intended to provide may be compromised. Unfortunately, the assumption that processes do not deviate from the protocol may not hold in open systems, where each process is under the control of a different entity. In fact, if the entities are selfish and they benefit from deviations, then they may change the protocol run by the processes. To avoid this problem, protocols must sustain cooperation, i.e., they must provide incentives that deny any benefits to the entities responsible for deviations.
One way of modelling selfish behaviour is to adopt the approach of Game Theory. In this approach, processes are seen as being under the control of rational agents that seek to maximize individual utility functions, interactions are modelled as games, and protocols correspond to strategies of the game that specify the actions taken at each point in time. The main goal is to devise equilibria protocols, i.e., protocols such that no agent increases its utility by causing a deviation. Equilibria protocols sustain cooperation, thus being extremely relevant to the development of dependable distributed systems.
In this work, we apply Game Theory to identify and analyse protocols that sustain cooperation in three fundamental distributed problems: (i) the problem gossip dissemination, (ii) the problem of pairwise exchanges of messages over links of a dynamic network, and (i) the problem of consensus with crash failures. Our main results identify necessary and sufficient conditions for devising equilibria protocols that solve the aforementioned problems. These results unveil the necessary and sufficient requirements for the construction of dependable distributed systems robust to selfish behaviour.
Blotter: Low Latency Transactions for Geo-Replicated Storage
Henrique Moniz, João Leitão, Ricardo Dias, Johannes Gehrke, Nuno Preguiça and Rodrigo Rodrigues
André Filipe Pessoa Negrão PhD Defense
20 de dezembro de 2016, pelas 10:30 horas, na Sala de Reuniões do Departamento de Engenharia informática (0.19), Pavilhão de informática II do IST
Xavier Vilaça PhD Defense
December 14th, 2016 at 10:30 am, in Anfiteatro PA-3 (-1 floor, Mathematics Building)
FastRank: Practical Lightweight Tolerance to Rational Behaviour in Edge Assisted Streaming
Xavier Vilaça, Luís Rodrigues, João Silva, Hugo Miranda
Accountability in Dynamic Networks
Xavier Vilaça, Luís Rodrigues
Automatic Generation of Policies to Support Elastic Scaling in Cloud Environments.
Richard Martinez, Luís Rodrigues
Redundancy Can Save You, When You Control It Smart
Pradeeban Kathiravelu, Luís Veiga