Eigen+: Memory Over-Subscription for Alibaba Cloud Databases

Memory is a critical bottleneck in cloud computing. While memory over-subscription enhances resource utilization by allowing more instances per machine, it increases the risk of Out of Memory (OOM) errors, potentially compromising service availability and violating Service Level Objectives (SLOs). Traditional methods predict memory utilization based on historical data and use bin packing algorithms for resource scheduling but often fail under high utilization due to inaccurate predictions. We observe that in cloud databases, the causes of OOM errors adhere to the Pareto Principle: the majority of OOM errors originate from a minority of instances. Leveraging this insight, we transform the complex prediction problem into a classification task by identifying and managing these problematic instances.