Cluster for beam dynamics error studies

The need for computing power for the dynamics of intense beams is growing. The error studies required to determine an accelerator sensitivity to imperfections call for calculations that are increasingly heavy (memory), long (CPU time) and numerous (statistics). Today, many laboratories are developing a more economical approach than supercomputers. It involves linking PCs via a network, whether dedicated or not. Together, they form a cluster. Depending on the software used to manage the cluster, parallel or multi-parameter computing can be distributed across this farm of PCs. The economic argument is obvious, and can be taken to extremes by using only machines already in the park. There's no need to spend a single penny on hardware. The use of office PCs at night and at weekends makes sense for multi-parameter calculations, which are made up of numerous processes, each relatively resource-efficient. As the accelerator error study falls into this category, the development of a cluster using the local LAN network was carried out at SACM. This report details the principles of the software layer that manages the cluster: client/server architecture, the TraceWin interface for client management