Logo

HPC @ Uni.lu

High Performance Computing in Luxembourg

Grid’5000 is a scientific instrument distributed in 10 sites (mainly in France) for research in large-scale parallel and distributed systems. It aims at providing a highly reconfigurable, controllable and monitorable experimental platform to its users.

The infrastructure has reached 1035 nodes and 7782 cores and all sites are connected to RENATER with a 10 GB/s link, except Reims (1 GB/s).

The University of Luxembourg, and its Granduc cluster are part of Grid’5000. Consequently, UL members can ask for an account, and use the 10 other sites of the Grid’5000.

G5K Account Creation

The below instructions are specific to the use of use Grid’5000.
To get an account on the other internal cluster of the UL, go on this page.

So to get an account on Grid’5000, you must:

  • read conscientiously the user charter. Your usage of Grid’5000 is automatically monitored, and must be conform to the user charter (total cpu time, number of reservations, user report, etc), otherwise your account may be suspended.
  • fill in the Account Request form.

Important At the Manager entry where you’re asked the Grid’5000 login of the person who’s responsible of the account you’re requesting, answer svarrett.

Specific tools

Grid’5000 provides a homogenous environment for large scale experiments:

  • OAR batch scheduler
  • Oargrid - conduct experiments on the grid (several clusters, several sites)
  • Kadeploy - operating system deployment, personalized environments
  • Kavlan - VLAN management on Grid’5000 nodes
  • API - allowing scripting and providing access to the grid functionalities

See the Grid’5000 web site to read more about the possibilities of the platform.

Luxembourg site hardware description

The cluster is organized as follows (click to enlarge):

As can be seen, it is composed of:

  • a front-end (virtual) server luxembourg.grid5000.fr that serves as an interface to the cluster, and which can be reached from a Grid’5000 access point.
  • an adminfront server, which hosts all of the cluster’s internal services in virtual machine instances
  • an NFS server, with a (raw) capacity of 28 TBs
  • 38 computing nodes
Important Sandybridge processors (petitprince-* nodes) should carry on 8 ops/cycle and support the AVX command set.

Interconnect

The interconnect is composed of:

  • one main router (Cisco 6500), which maintains the link with the French G5K network
  • the 10GbE network (using Cisco Nexus 5000 and Force10 switches)

Storage / Cluster File System

The Luxembourg site of Grid’5000 hosts 2 storage servers:

  • the first server is the NFS, which is attached to a Dell MD1000 disk enclosure. This enclosure contains 14 SATA disks, configured in RAID50, for a raw capacity of 28TB.
  • the second server is dedicated to Storage5k (storage on demand), and contains 6 SAS disks (400GB, 10K rpm), for a raw capacity of 2.4TB.