Storage Workload Isolation via Tier Warming: How Models Can Help

Ji introduced his team’s workload prediction model and scheduling method for the purpose of isolating the performance of user traffic from system background work. He first introduced the tiered storage system and then explained that scheduling background work may flush out the user working set in the fast tier and therefore impact the performance of user traffic. Then he showed how their Markovian-based model can predict the high intensity of future-user workload and proactively warm up the fast-tier with the user working set to prevent performance disaster. Finally, he showed performance comparison results between their proposed method and the state-of-the-art industry solution, the feedback method.