The computing platformΒΆ

The computing platform accessible by the job scheduler is composed of Linux computing servers. It is composed of three main use cases (please find some use examples in the Types of jobs page):

The HTC platform (High-Throughput Computing)
is suitable for running most traditional HEP mono or multi-core applications: it accounts for most of the computing power made available at CC-IN2P3.
The HPC platform (High-Performance Computing)
of smaller capacity, is designed to accommodate parallel calculations. It is composed of a set of servers connected in Infiniband, which allows an effective communication inter-servers via the use of MPI libraries.
The GPU platform
is composed of a group of servers equipped with graphic cards to accommodate vector calculation applications.

Note

To know the technical characteristics of the compute servers, please refer to the page Compute servers configuration.

The scheduler is the only entry point common to all users to submit jobs on the computing platform. Its role is to receive jobs submitted by users, to schedule them and submit them for execution on an appropriate and available computing server.

The main goal is to use the computing resources (memory, disk space, CPU) in the most efficient way possible. The sharing of all resources for all users allows optimal use of the entire computing platform.

In general, a job is a task (or set of tasks) that the user wants to run on the servers of the computing platform. This task can be an executable file, a set of commands, a script, and so on. A job can be developed and tested on interactive servers before being massively submitted to the computing platform.

A job is always submitted on an execution queue. Each execution queue has default values for disk space, CPU time and memory. There are several queues that, needing a lot of resources, are restricted. In this case, the user will need explicit access authorisation (see the restricted queues FAQ).

All queues allow the simultaneous execution of many jobs. The system always tries to launch new jobs in a least loaded and most appropriate queue.

Please find below the documentation covering the job submission and management.