Service Profiling Based Management and Scheduling in Kubernetes

Learn about an open-source solution for efficient resource management and scheduling in Kubernetes through this 21-minute conference talk from ByteDance engineers. Discover how to construct workload-specific resource profiles using historical utilization patterns to optimize resource allocation. Explore the implementation of Service Profiling Description (SPD), a custom resource that maps workloads to their resource usage patterns across deployments and stateful sets. Understand how various metrics including CPU, disk I/O, and network I/O are collected and leveraged to inform intelligent scheduling decisions. See real-world applications of this solution in large-scale Kubernetes clusters, handling diverse workload requirements from dedicated NUMA nodes to shared resource environments.