This talk presents an implementation of overload management for the ATLAS soft real-time scheduler. In the scope of my thesis, I examined five different strategies for distributing a required cutback among jobs of an overloaded schedule in ATLAS. Information about the current load situation is communicated from the kernel to the runtime in userland. In userland, a load manager was implemented, that uses this information in conjunction with functional alternatives and iterative refinement functions to adapt to varying load situations.
Managing Overload with ATLAS Real-Time Scheduling
Verteidigung der Diplomarbeit, abweichender Raum: APB 2101