Grid5000

Grid5000
Name	Grid5000
Established	2003
Type	Research infrastructure
Fields	High-performance computing, Distributed systems, Cloud computing
Country	France

Grid5000

Grid5000 is a French experimental distributed computing testbed designed for large-scale research in high-performance computing, computer networking, parallel computing, and cloud computing. It provides reproducible, controllable environments for experiments involving clusters, networking, virtualization, and middleware used by researchers from institutions such as Commissariat à l'énergie atomique et aux énergies alternatives, CNRS, INRIA, Université Paris-Saclay, and international partners like DEISA and PRACE. The infrastructure emphasizes instrumented platforms, repeatability, and collaboration among European and global projects including European Grid Infrastructure and OpenStack initiatives.

Overview

Grid5000 offers an integrated facility that combines bare-metal clusters, programmable network fabrics, and virtualized resources to support experiments in distributed systems, middleware, and performance evaluation with traceability for publications at venues such as ACM SIGCOMM, USENIX, IEEE INFOCOM, and EuroSys. The project supports users from research organizations like INRIA Saclay, École Normale Supérieure, Sorbonne Université, Université Grenoble Alpes, and international collaborations with entities like NCSA and TACC. Core capabilities include automated deployment, resource reservation, and instrumentation for experiments cited alongside works from groups at MIT CSAIL, Stanford University, UC Berkeley, and ETH Zurich.

Architecture and Infrastructure

The testbed architecture comprises clusters of commodity servers, programmable switches, and dedicated storage arrays interconnected by high-speed links and managed by orchestration software influenced by projects such as Kubernetes, OpenStack, Xen Project, and KVM. Its facility management integrates identity and accounting systems compatible with standards from GLUE schema, LDAP, and authorization frameworks used by EUDAT and EGI. Hardware across sites includes servers from vendors like Dell EMC, Hewlett Packard Enterprise, and networking gear from Juniper Networks and Cisco Systems, while monitoring stacks incorporate tools inspired by Nagios, Prometheus, and Zabbix.

Research and Applications

Researchers use the platform to study fault tolerance algorithms tested against workloads from projects affiliated with LHCb, CERN, and simulations aligning with results reported at SC (Supercomputing Conference), ICSE, and SOSP. Applications span distributed file systems compared to Ceph and Lustre, container orchestration research tied to Docker and Kubernetes, and network function virtualization evaluations referencing NFV frameworks used by ETSI. Work on energy-aware scheduling references studies from European Commission programs and collaborations with sustainability initiatives at CEA and CNRS labs.

History and Development

Initiated in the early 2000s, the project evolved through phases influenced by European research programs including FP6, FP7, and Horizon 2020, with contributions from French institutions such as INRIA, CNRS, CEA, and universities like Université de Lorraine and Université de Rennes 1. Milestones include expansions concurrent with grid computing efforts like gLite and federations engaging with infrastructures such as EGI and partnerships connected to projects at LIP6 and IRIT. The platform’s software stack and management systems drew on antecedent research from Grid Computing efforts and contemporary practices at laboratories like Laboratoire de l'Informatique du Parallélisme.

Governance and Funding

Governance involves steering committees and technical boards comprising representatives from national research organizations including CNRS, INRIA, Ministère de l'Enseignement Supérieur et de la Recherche, and partner universities like Université de Lille and Université de Strasbourg. Funding has been provided through grants from European programs such as Horizon 2020, national funding agencies like ANR, and institutional support from CEA and regional councils including Région Île-de-France. Collaborative agreements and consortium models echo frameworks used by infrastructures like PRACE and European Grid Infrastructure.

Deployment Sites and Testbeds

The infrastructure is geographically distributed across multiple French sites hosted at laboratories and universities including Université de Lyon, Université Grenoble Alpes, Université de Rennes 1, Université de Strasbourg, and Université Paul Sabatier (Toulouse III), enabling multi-site experiments akin to federated testbeds such as PlanetLab and Emulab. Each site features homogeneous clusters, management nodes, and network slices interoperable with external testbeds like Grid'5000 partners, while experimental topologies often mirror configurations studied at Microsoft Research, IBM Research, and Google Research.

Performance and Evaluation

Performance evaluations on the platform leverage benchmarking suites and methodologies comparable to those used at Top500 rankings, studies published in IEEE Transactions on Parallel and Distributed Systems, and reproducibility standards promoted by ACM Reproducibility. Metrics include network latency, throughput, job scheduling efficiency, virtualization overhead examined against Xen and KVM baselines, and storage I/O compared with Lustre and Ceph implementations, with results informing best practices at conferences like SC and EuroSys.

Category:Research infrastructures