abstract 2016

The abstracts below are for PhD students talks. Each presentation will be of 10 minutes max followed by 5 minutes of questions while the next speaker sets up.

Christophe Bacara (Université Lille 1 - Stormshield)

Use case : Alarm telemetry within a large set of IPS

This presentation will be about the first use-case of my PhD thesis. In order to gather informations about security alarms occuring on Stormshield's UTMs, we need to be able to send a request to a set of UTMs and aggregate the responses' data. The confidentiality of gathered data is essential, as it can leak informations about an organisation's internal IT infrastructure. Furthermore, there is a few constraints. For example, we may need to request data for a certain amount of time, which can be greater than the logs storage capacities : thus, we must be able to ask an UTM to store additionnal informations in order to respond to the requests. Another constraint is the ability to request some kind of informations that weren't available at deployment time, according to the UTM's firmware lifecycle and new attacks signatures.

Simon Bouget (IRISA / Inria Rennes)

Topology composition in distributed networks

Designing systems-of-system is far from being an easy task, in particular with the advent of the Internet of Things (IoT) that involves bringing together heterogeneous distributed entities to collaborate towards a common global objective. Developing such systems involves being able to specify distributed systems topologies that reach a new level of complexity. To deal with this complexity, we introduce a new programming model, that allows to easily combine different basic shapes (ring, grid, etc.) so that developers can define complex topologies as an assemblage of simpler blocks.

Nadir Cherifi (Lille 1 / Worldline)

An Introduction to Energy Profiling of Small Connected Objects

Surrounding autonomous embedded devices are in constant expansion. By the rise of Internet of Things (IoT), we find today more and more objects around us, deployed in multiple real-world applications of our everyday life.

Predictions reveal that a huge number of connected objects are expected to be deployed over the coming years; many of them will be energy-dependent, and constrained by a battery. Thus, energy represents a critical resource that limits the deployment at a very large scale of these smart connected devices. This energy constraint greatly complicates the development of the software embedded by these objects. Therefore, the ability to measure, track and finely profile the power consumption of such objects, and correlate it with the application driving the device functionalities, is a big challenge to improve the development of IoT embedded applications.

In this presentation, we aim to give an introduction to the energy consumption issue regarding the software development of energy constrained objects. Specially, we’ll describe the main limitations of existing energy profiling approaches, and our work in progress to overcome them.

Hadrien Croubois (ENS Lyon)

Toward an autonomic approach of workflows distribution on cloud

Advances in distributed systems technologies require a constant rethinking of previous deployments methods. Development of the Cloud paradigm is symptomatic of a broader tendency toward more dynamicity in the management of tasks and resources. However, the scientific computing still mostly uses old paradigms when deploying complex workflows. Our focus is therefore to propose a solution that will make the link between the needs of user in terms of scientific computation and the features

offered by cloud providers. After having modelled both those needs and features, we describe the different mechanisms which are part of what should allow for an autonomous platform dedicated to collaborative scientific computing.

Stéphane Delbruel (IRISA Rennes)

Mignon: a Fast Decentralized Content Consumption Estimation in Large-Scale Distributed Systems

Although many fully decentralized content distribution systems have been proposed, they often lack key capabilities that make them difficult to deploy and use in practice. Here, we look at the particular problem of content consumption prediction. We propose a novel, fully decentralized protocol that uses the tags attached by users to on-line content, and exploits the properties of self-organizing kNN overlays to rapidly estimate the potential of a particular content without explicit aggregation.

Boris Djomgwe Teabe (IRIT, SEPIA)

Application-specific quantum for multi-core platform scheduler

Scheduling has a significant influence on application performance. Deciding on a quantum length can be very complex, especially when concurrent applications have various characteristics. This is particularly the case on virtualized systems in cloud computing environments where virtual machines from different users are colocated on the same physical machine. We claim that in a multi-core virtualized platform, different quantum sizes should be associated with different application types. We applied this principle in a new scheduler called AQL Sched. The latter dynamically associates an application type with each virtual CPU (vCPU). We identified 5 main application types and experimentally found the best quantum length for each of them. Subsequently vCPUs of the same type are scheduled within a dedicated pool of physical CPUs (pCPUs), with the quantum length associated with this type of vCPU. Therefore, each vCPU is scheduled on a pCPU with the best quantum.

Lyes Hamidouche (UPMC, LIP6)

Toward efficient data delivery over WiFi Networks using D2D connections

Conventions and exhibitions are events where plenty of people gather at the same place for a common interest. The mobile applications that are used during these events consume important amounts of data, like videos, pictures, documents, web pages, etc. The Client/Server model is still commonly used for this kind of software, however it is not scalable, there is an important amount of the same data to download on the numerous nodes. As mobile devices have the ability to establish Device­to­Device (D2D) connections to exchange data, we have designed solutions based on this model of communication to provide an efficient downloading of data through WiFi Networks. The D2D model popularity is increasing, due to the constant improvement of the performance of the mobile nodes on the market. Our approach consists in using D2D connections over a wifi network to forward the common data that the nodes download in order to provide an efficient access.

Mahsa Najafzadeh (LIP6)

The analysis and co-design of weakly consistent programs

Designers of a replicated database face a vexing choice between strong consistency, which ensures certain application invariants but is slow and fragile, and asynchronous replication, which is highly available and responsive, but exposes the programmer to unfamiliar behaviours. To bypass this conundrum, recent research has studied hybrid consistency models, in which updates are asynchronous by default, but synchronisation is available upon request. To help programmers exploit hybrid consistency, we propose the first static analysis tool for proving integrity invariants of applications using databases with hybrid consistency models. This allows a programmer to find minimal consistency guarantees sufficient for application correctness.

Vlad-Tiberiu Nitu (Toulouse INPT)

DIVA: DIstributed Virtualization Architecture

Virtual machine (VM) consolidation is limited by today’s virtualization systems which are capable to start a VM exclusively on a server that is able to provide the entire amount of resource needed by that VM. This constraint causes resource fragmentation in the cloud, leading to a significant amount of resource waste. We present DIVA (DIstributed Virtualization Architecture), a new virtualization architecture which aims to remove the physical boundaries between resources in a datacenter. DIVA is a cooperative two-layer distributed system involving both a distributed hypervisor and a distributed operating system (for VMs). These two components enable a VM to use, at the same time, resources provided by multiple servers.

Pierre-Louis Roman (Université de Rennes 1 - IRISA)

GPS: Priority Gossip for Differential Consistency

Consistency has gained a lot of focus since the publication of the CAP theorem. To circumvent the impossibility of having consistency, availability, and partition tolerance, researchers have proposed solutions offering various trade-offs between these three properties. These trade-offs tend to be uniformly applied to all the nodes in a system, but there is a growing demand for differentiated requirements, particularly for consistency.

We address this demand with Gossip Primary-Secondary, a novel epidemic dissemination protocol that offers differentiated delivery latency and consistency by working in pair with an update-consistent protocol. We demonstrate through analysis and simulations with a network of one million nodes that the delivery speed can be analytically computed and verified through experiments. We measure the consistency experienced by nodes with a novel consistency metric that we formally define with its associated data type.

Alejandro Tomsic (LIP6)

Physics: Efficient consistent snapshots for scalable snapshot isolation

Non-Monotonic Snapshot Isolation (NMSI), a variant of the widely deployed Snapshot Isolation (SI), aims at improving scalability by relaxing snapshots. In contrast to SI, NMSI snapshots are causally consistent, which allows for more parallelism and a reduced abort rate.

This work documents the design of PhysiCS, a transactional protocol implementing NMSI in a partitioned data store. It is the first protocol to rely on a single scalar taken from a physical clock for tracking causal dependencies and building causally consistent snapshots. Its commit protocol ensures atomicity and the absence of write-write conflicts.

We argue that PhysiCS approach increases concurrency and reduces abort rate and metadata overhead as compared to state-of-art systems.

Gauthier Vauron (INRIA/LIP6)

Everything you want to know about virtualization on large multicores

Cloud providers are now proposing large virtual machines with several tens of cores. Since such virtual machines often relies on complex hardware architectures, new challenges arise for hypervisors to achieve

good performances. This talk will present some of these problems and why they happen.

Lofti Zaouche (Université de Technologie de Compiègne)

Flying Ad-Hoc Networks (FANETs): An application

In the time where Unmanned Aerial Vehicles (UAVs) knows a large expansion in public use, many applications emerged, such as overfly an inaccessible location and take pictures, or transporting objects. The research community try to exploit these new technology to propose new solutions. During our research work, we studied the moving ability of UAVs to construct a dynamic network and evaluated the performance of a such network. We studied the case of follow and film a Moving Point of Interest (MPI) and transmit the stream to a Ground Station (GS). This application can be used in a rescue operation, or filming a sport event.