Section: New Results

Experimentation

This section covers our work on experimentation on testbeds (mainly Grid'5000), on emulation (mainly around the Distem emulator), and on Reproducible Research.

Grid'5000 Design and Evolutions

Participants : Benjamin Berard [SED] , Luke Bertot, Alexandre Merlin, Lucas Nussbaum [contact] , Nicolas Perrin, Patrice Ringot [SISR LORIA] , Teddy Valette [SED] .

The team was again heavily involved in the evolutions and the governance of the Grid'5000 testbed.

Technical team management. Since the beginning of 2017, Lucas Nussbaum serves as the directeur technique (CTO) of Grid'5000 in charge of managing the global technical team (10 FTE). He is also a member of the Bureau of the GIS Grid'5000.

SILECS project. We are also heavily involved in the ongoing SILECS project, that aims to create a new infrastructure on top of the foundations of Grid'5000 and FIT in order to meet the experimental research needs of the distributed computing and networking communities.

SLICES ESFRI proposal. At the European level, we are involved in a ESFRI proposal submission. We submitted a Design Study project in November 2019, and are in the final stages of submitting the ESFRI proposal itself in early 2020.

TILECS workshop. We participated in the organization of the TILECS workshop. TILECS (Towards an Infrastructure for Large-Scale Experimental Computer Science, https://www.silecs.net/tilecs-2019/) gathered about 80 members (mostly faculty) of the testbeds designers and users community in France, to discuss the future plans for research infrastructures in the networking and distributed computing fields. During that workshop, Lucas Nussbaum presented Grid'5000 [32].

Group storage. A technical contribution from the team is the addition of a group storage service that allows groups of users to share data, with improved security and performance compared to what was previously available.

Support for Debian 10. Another notable technical contribution from the team is the work of Teddy Valette on supporting Debian 10 in the set of Grid'5000 system environments made available to users.

New clusters available in Nancy: graffiti, gros, grue. Finally, the team was also heavily involved in the purchase and installation of several new clusters in the Nancy site, gathering funding from CPER LCHN, CPER Entreprises, MULTISPEECH team, LARSEN team. This greatly increases the resources available locally, both for GPUs (graffiti and grue), and for large-scale experiments (gros).

Involvement in the Fed4FIRE Testbeds Federation

Participants : Luke Bertot, Lucas Nussbaum [contact] .

In the context of the Fed4FIRE+ project (section 9.3.1.1), Grid'5000 was officially added to the Fed4FIRE federation at the beginning of 2019. In 2019, we implemented on-demand stitching between Grid'5000 experiments and other testbeds of the federation (through VLANs provided by GEANT and RENATER), allowing experiments that combine resources from Grid'5000 and other testbeds [27]. We are also improving our implementation of an SFA Aggregate Manager in order to allow the use of Grid'5000 through Fed4FIRE tools, such as the jFed GUI.

We also worked on the issue of classifying and presenting the set of testbeds available in the federation. This was the subject of a presentation at the GEFI collaboration workshop [31].

I/O Emulation Support in Distem

Participants : Alexandre Merlin, Abdulqawi Saif, Lucas Nussbaum [contact] .

We finished the work on adding I/O emulation support in Distem, in order to experiment how Big Data solution can handle degraded situations [22].

Distributing Connectivity Management in Cloud-Edge infrastructures

Participant : Lucas Nussbaum [contact] .

In the context of David Espinel's PhD (CIFRE Orange, co-supervised with Adrien Lebre and Abdelhadi Chari), we worked on distributing connectivity management in Cloud-Edge infrastructures [38]. The classic approach of deploying large data centers to provide Cloud services is being challenged by the emerging needs of Internet of Things applications, Network Function Virtualization services or Mobile edge computing. A massively distributed Cloud-Edge architecture could better fit the requirements and constraints of these new trends by deploying on-demand Infrastructure as a Service in different locations of the Internet backbone (i.e, network point of presences). A key requirement in this context is the establishment of connectivity among several virtual infrastructure managers in charge of operating each site. In this work, we analyzed the requirements and challenges raised by the inter-site connectivity management in a Cloud-Edge infrastructure.

NDN Experimentation

Participants : Thibault Cholez [contact] , Xavier Marchal, Olivier Festor.

While ICN is a promising technology, we currently lack experiments carrying real user traffic. This also highlights the difficulty of making the link between the new NDN world and the current IP world. To address this issue, we designed and implemented an HTTP/NDN gateway (composed of ingress and egress gateways) that can seamlessly transport the traffic of regular web users over an NDN island, making them benefit from the good properties of the protocol to deliver content (request mutualization, caching, etc.). The gateway itself is part of a wider architecture that aims to use NFV to deploy NDN and benefit from its orchestration capability to address performance and security issues inherent to new network architectures.

To validate the whole architecture, a testbed involving real users was made. The gateway was used by dozens of users for a few weeks to prove that running a NDN network over NFV is a viable solution to address the transition between both worlds. Users accessed many websites through the NDN network in a very satisfying way. The results have been published in IEEE Communications Magazine  [5].