Calculer sur la grille avec CiGri

Cigri in a (very small) nutshell

CiGri is a lightweight grid dedicated to process bag-of-tasks applications that may result in a very large number of small jobs, like multi-parametric applications (where you have to run the same program on a large set of parameters and get one result per parameter). It’s particularity is to run on unused cpu cycles of a set of supercomputers in a computing center, in a “best-effort” way. But it may also be used to manage a large set of tasks to run on a particular cluster even in a non-besteffort way (large set of tasks usualy overload the local batch scheduler of a cluster; the grid is made to deal with this problem).

Best-effort jobs are zero priority jobs that may be killed without notice if a normal job needs the resources. It’s a way to compute transparently on clusters that are not yours, only when cpus are idle. CiGri is a useful tool to manage such jobs because you may have to submit again and again the same job until it is not killed. We call this re-submission and CiGri does this automatically.

Requirements to understand CiGri

To compute efficiently on the local grid and take Cigri in hand, you have to know how to:

Then, you can read the following tutorials, each based on a specific way to set up properly your software environments across the targeted clusters, to learn how CiGri works: