Explore the concept of metastable failures in large systems through this 38-minute conference talk from SREcon24 Americas. Delve into the characteristics of self-perpetuating performance failures and their positive feedback loops that keep systems in a degraded state. Learn why operating in a metastable vulnerable state is sometimes unavoidable and how overprovisioning to prevent it can lead to poor resource utilization and high costs. Discover three key strategies for mitigating the risk of metastable failures: understanding environments, algorithms, and workloads; implementing metastable failure trigger-resistant design; and protecting vulnerable components. Gain valuable insights from speaker Aleksey Charapko of the University of New Hampshire on effectively managing and mitigating risks in complex systems.
Overview
Syllabus
SREcon24 Americas - It Is OK to Be Metastable
Taught by
USENIX