In total this academical year, five Master Theses and one internship have been conducted in collaboration with 3 different academical institutions in Belgium: Université Libre de Bruxelles (ULB), Université de Liège (ULg) and Université Catholique de Louvain (UCL). Let’s here quickly review their goals and contributions.
Model-Driven Approach for the Generation of a Distributed Choreography from a BPMN Orchestration Process.
In this Master Thesis we propose to define a model of transformation which aims at decomposing a business process into a set of distributed choregraphies. The goal is then to execute the business process in a fully distributed manner on an Event-Driven Architecture. The advantage of this is to remove the process execution engine as a bottleneck and provide scalability in the execution of the processes.
Link to the Thesis of M. Mounir Chaal (ULg).
Recommendation Engines: a Trust-based approach for Real-Time Social Network Services.
Recommendation engines are used in social networks but also in other online services such as shopping sites. These aim at providing the user information which is relevant to her based on the online profile and activity. This can in practice for example consist in the “people you may know” or “articles you would be interested in” features of such sites. In this Master Thesis we propose a recommendation algorithm based on the measure of trust between members of a social graph. We also propose a measure for trust in such social networks.
This Master has led to a number of 2 scientific publications:
- “Towards trust inference from bipartite social networks” at the 2nd ACM SIGMOD workshop for Databases and Social Networks, DBSocial, Scottsdale Arizona, USA
- “Trust-based recommendation: An empirical Analysis” at the Sixth ACM workshop on Social Network Mining and Analysis, SNA-KDD, Beijing China
Link to the Thesis of Daire O’Doherty (UCL).
Cloud Service Managers and Open Standards.
As Cloud infrastructures and services become more and more popular and the available offers keep on getting affordable, most of these Cloud providers provide closed interfaces to the resources. This generally leads to lack of portability and interoperability between different cloud based platforms. OCCI is an initiative which aims at providing open specifications for managing Cloud resources. However, it only covers the infrastructure (IAAS) management level of Cloud services. In this Thesis we propose an extension of the OCCI model to cover to the service level.
Link to the Thesis of Hervé Koener (ULg).
Study of Distributed Data Grid Frameworks as Elastic Storage for the Cloud.
Cloud computing has enabled the elastic scaling of applications by allowing to dynamically add new computing instances to the existing cluster. This has lead to the development of new data structures and databases model such as NoSQL which have allowed to scale the data of such applications in a cloud environment (Euranova R&D has led many Master Theses in this particular domain last year). In this internship we studied the relevance of Data Grid structures as elastic storage back-end for applications in the Cloud environment.
Link to the internship report of Stephanie Glele (ULg).
Real-time allocation of Virtual Machines in the Cloud.
Though the Cloud has enabled the dynamic allocation of resources on a pay-per-use basis, allowing to elastically scale out applications, there still remains the problem of latency from the moment a new instance is ordered and the moment it is up and running. This latency can have a bad impact on the performance of services which can not afford (for instance real-time voice translating services, MMOG servers, …). Most of the current public cloud providers offer little guarantee over this latency. In this Thesis we propose a model of VM pooling as to ensure that a required machine instance is immediately up and running at the time it is requested which is more affordable than over-provisioning.
Link to the Thesis of Benjamin Hesmans (UCL).
Elastic Block Devices and Allocation of VMs Close to the Data.
Open Nebula is one popular Open Source Cloud platform which allows users to build and manage their own Cloud infrastructure on their hardware. In its current implementation Open Nebula only stores the VM images in a single directory on a same host. All the images need to be copied on the host where the VM will be running prior to its deployment. The goals of this Thesis are first to implement a distributed and replicated image storage inside the current Open Nebula implementation, then to extend the scheduling model to take into account the colocalization and replication of the images and the hosts available for deployment.
Link to the Thesis of Jorick Flabat (UCL).
One ship leaves, another arrives.
As ever, EURA NOVA is renewing its commitment towards academic research and will be in charge of another round of Master Theses. This year the contributions will especially be covering the domains of graph related storage & processing and elastic message queues.