Integrated resource allocation in shared datacenters

Software-defined datacenters consolidate discrete hardware units into pools of abstract resource types that are dynamically allocated to the clients sharing the infrastructure. The allocation needs to be both fair to the clients and also make efficient use of the system resources. Unlike approaches that allocate individual resources independently, we propose a model for making integrated allocation decisions across multiple resource types, so as to balance fairness and system throughput in a heterogeneous-resource setting.