Running low-latency workloads on Kubernetes: Lessons learned from SpiceDB - Sohan Maheshwar

This talk focuses on optimizing Kubernetes for performance-sensitive workloads, highlighting the challenges such as unpredictable scheduling, resource contention, and noisy neighbor effects. The speaker, Sohan Maheshwar, draws from his experiences running SpiceDB, a low-latency authorization system, demonstrating how to achieve 5ms p95 at 1 million requests per second. Key topics include understanding Kubernetes defaults, implementing advanced resource management strategies, and techniques for scaling applications for optimal performance. Attendees will learn practical steps to mitigate latency issues and ensure their systems operate efficiently under high demands.