GRIF, a Distributed Grid Resource

GRIF is a joint initiative from 5 CNRS/IN2P3 laboratories and CEA/IRFU (ex CEA/DAPNIA), all located in Paris region, in order to build a major grid resource. GRIF is seen as one grid site but its resources and its technical team are spread over 6 geographical locations.

Driven by the needs of LHC experiences, which rely on grid for their computing, GRIF has always been opened to many communities not related to High Energy Physics, like biomed, earth sciences, chemical chemistry, astro-physics... In 2007, GRIF started a partnership with Institut des Systèmes Complexes Paris Ile-de-France to provide them a significant computing power for their needs, around 3 MSI2K.

GRIF faced many challenges, trying to build this distributed resource. Probably, the most important one was to build a unified technical team with people spread in the different laboratories and not necessarily dedicated to GRIF management. This technical team is currently made of around 20 people, representing 10 FTEs. This is the core asset of GRIF.

To leverage administration of this resource, GRIF chose a configuration and management tool, Quattor. This tool is unique in its ability to share management of the whole resource in a controlled manner, ensuring consistency of the global configuration and still preserving necessary autonomy of each GRIF site.

The other important challenge was to build a network infrastructure allowing seamless access to data from any computing resource located in any GRIF site. GRIF started with general-purpose RENATER connection between each of its sites (1 Gb/s) and is now moving to a private 10 Gb/s private network, GRIFOPN, based on dark optical fibres provided by RENATER. This will be a major step forward for hiding resource distribution to users.

Beginning of 2008, GRIF is a large grid site, with consolidated computing resources around 5 MSI2K corresponding to around 20 TeraFlops and storage resources around 500 TB of disk. In addition to these, GRIF is also operating several general-purpose grid services used by many other sites, like a top-level BDII, a WMS (job submission) or a VOMS server (VO management).