Radar: Reliable Resource Scheduling for Composable/Disaggregated Data Centers

Abstract

Hardware disaggregation decouples resources (e.g., processors and memory) from monolithic servers, potentially improving service reliability. However, from another perspective, directly exposing resource modules to a shared network may adversely affect service reliability. In this paper, we study a reliable resource allocation problem in disaggregated DCs (DDCs), considering network impact and different disaggregation scales. We provide a mixed-integer linear programming formulation and a resource allocation framework named Radar for this problem. Numerical results demonstrate that the benefits of hardware disaggregation may be adversely affected by an imperfect network. It also shows that both the hardware backup and a proposed migration-based restoration can be applied to overcome this potential adverse effect.