Research

The Magellan project-team aims to enable the creation of efficient, robust, environment-friendly and rigorously-evaluated decentralized computing infrastructures. It focuses its research efforts on reliable decentralized computing infrastructures, reliable decentralized application runtimes, distributed infrastructure frugality, and evaluation methodologies and tools.

Reliable decentralized computing infrastructures

Decentralized computing infrastructures constitute the foundation which supports virtualized computing, storage and networking services upon which a wide range of sophisticated geo-distributed applications may be developed and executed. They should be scalable according to the number of nodes and their broad geographical distribution, reliable in the face of fluctuating node and network performance and availability, and easy to maintain and to operate.

Within this axis we focus on the design of scalable and reliable virtualized execution platforms, application lifecycle management, federated infrastructure design, multi-tenancy, networking services and data management. A long-term goal is to enable the development of generic fog computing platforms which may eventually become public services operated by local authorities (e.g., cities, regions) to serve a large number of requirements stemming from their own usage as well as those from local companies and citizens. Beyond the performance- and efficiency-related motivations for such platforms (and their associated research challenges regarding scalability, stability, usability, resource management, data management etc.), we expect that sovereignty requirements regarding application’s data and the infrastructures which manage them are going to become increasingly important from a social as well as scientific point of view.

Reliable decentralized application runtimes

Decentralized application runtimes should support developing, deploying and managing complex applications in a simple and natural way, while hiding the complexities of operating the underlying infrastructure. The runtimes should support automatically managing the quality requirements of applications and responding to changes in operating conditions, while cost-effectively provisioning resources from the decentralized infrastructure.

The ongoing work aims to deliver a set of middleware environments to facilitate the development and operation of decentralized applications which exploit the underlying decentralized infrastructures to the fullest extent. The long-term objective is to make the development and operation of these applications no more complex than the development of cloud-based applications is today. We expect to invest our efforts on middlewares which support programming models such as data stream processing and function-as-a-service computing, where the main research challenges deal with automated application management, resource provisioning, auto-scaling, resource and data sharing within and across data centers, handling data bursts and mitigating stragglers.

Distributed infrastructure frugality

The Magellan team aims to reduce as much as possible the environmental impact of large-scale fog/cloud platforms. Understanding and reducing the environmental impact of the full life-cycle of our hardware and software resources is a difficult challenge. For instance, the manufacturing phase is dominating in the environmental impact of many ICT devices. Reducing their energy consumption during their use phase therefore does not necessarily guarantee lower environmental impacts, and increasing the device lifetime is crucial. As many such devices are deeply integrated within complex Cloud infrastructures, reducing the energy consumption of one part may in fact increase it for another part. In this context, we plan to further characterize the energy consumption of digital infrastructures, and to leverage the collected information to improve the efficiency of these infrastructures.

Evaluation methodologies and tools

The Magellan team aims to help professionalize the scientific evaluations in our research domain. We plan to push further our expertise in experimentation and simulation by combining these approaches. Building a mathematical surrogate of a given system (i.e., a simulation model of the real system) constitutes a highly interesting challenge: the assessment and improvement of the model provides a deeper understanding on the system, while the resulting surrogate enables accurate performance predictions without any execution on real platforms. Co-developing the real system and its surrogate model in the long term is a very promising direction to efficiently design, build and operate the complex systems that constitute the modern large-scale distributed infrastructures.

Comments are closed.