Grid’5000

Grid’5000 is a large-scale and versatile testbed for Cloud, HPC, Big Data, networking, and AI. It is composed of 8 sites (located in France and Luxembourg), and more of 800 servers than can be reserved at the bare metal level. The sites are connected together with a 10-Gbps dedicated backbone.

Map of Grid’5000 (left). Some Grid’5000 resources (right)

Resources include, as of early 2020: (up-to-date information is available at https://www.grid5000.fr/w/Hardware)

  • 34 clusters
  • 829 nodes (servers)
  • 14828 CPU cores
  • 156 GPUs
  • 10 Gb Ethernet, 25 Gb Ethernet, Omni-Path, Infiniband, etc..

The main features of the testbed are:

  • Reservation of resources at the bare metal level, with the ability to deploy custom environments (Debian is provided, Centos and Ubuntu will be provided soon). Experimenters can typically reserve a set of nodes (possibly large)  and configure their own software stacks on top of them. Tools are provided to help the deployment of some software stacks, such as OpenStack.
  • Network reconfiguration at the VLAN level to isolate experiments from the rest of the testbed, or build custom topologies.
  • Advanced monitoring capabilities, especially for power measurements. Using specialized devices, experimenters can monitor the power consumption of the nodes they reserved.
  • Support for Big Data experiments, with large storage spaces available for medium-term storage inside the testbed, and nodes with several disks (HDDs and SSDs).

Specific features for particular kinds of experiments

Wired networking experiments & SDN experiments

  • Dedicated 10 Gbps backbone between sites, isolated from the public Internet, with isolation between users (VLANs)
  • Nodes with up to 5 network interfaces (4x10G, 1x1G) ; many nodes with two connected 10G network interfaces
  • Bare-metal server access with support for running DPDK and OpenVSwitch

Cloud experiments

  • Support for deploying OpenStack clouds in the context of the user experiment (fully controlled by the experimenter, and possibly modified by the experimenter), using ENOS
  • Other software stacks are supported by the community (e.g. Kubernetes)
  • Basically any complex software stack can be installed on Grid’5000 thanks to the bare-metal deployment and network reconfiguration features

Big Data experiments

  • Nodes with up to five disks (HDDs or SSDs)
  • Disks on nodes can be reserved outside of nodes reservations
  • Total storage space available on nodes: 1100 TB
  • Shared storage space available on the testbed (Ceph clusters and large NFS servers)
  • GPU nodes for Big Data processing

Contact

Email: Lucas Nussbaum lucas.nussbaum@loria.fr

More details

Start typing and press Enter to search