Two formulations for a grid cyclic scheduling problem

Grid cyclic scheduling : the EGEE case Grid computing is now a major way to tackle large, possibly distributed computations. The european EGEE project [3] is a nice example of such systems. The EGEE grid is composed of many sites all around Europe and abroad. Each site disposes of several hundreds of computer units. Its first aim is to provide enough computation and storage resources for scientists, and especially for the physicists that will exploit the data of the LHC. Each experiment with the collider will provide an enormeous amount of data. It is also used for other applications, for instance concerning biology or health. The major issue when managing such enormous systems is the load balancing, while keeping consistency. EGEE is currently managed in a centralized way. This management raises complicated problems that are not within the scope of this presentation, which focuses on the management of one site. Each site is a parallel system. We will suppose here this system is a large cluster (that is the case for the site of the LPC, a physics laboratory in Clermont Ferrand, France. See [2]). The cluster receives a large number of tasks. These tasks are usually independent. They come from different users, and have different characteristics (length, periodicity, memory requirements,...). During these months before the use of the LHC as an operational system, many simulations are done to verify if the grid, and the individual clusters, can bear the load. Typically, a LHC experiment will generate a flow of tasks (50 per second) with a large variety of durations (some with a few seconds of duration, some with a few hours). However, other types of tasks will have to be dealt with in the same time.