Write-Through Object Cache for Fleets of Global Edge Clusters

This conference talk by Ryan Beisner from NVIDIA explores a system designed to accelerate and optimize S3 object storage performance across global edge Kubernetes clusters. Learn about an innovative write-through object cache solution specifically developed for GPU-focused workloads like cloud gaming and AI inference jobs that face challenges with egress costs and latency. Discover the architecture built on Kubernetes that leverages AIStore with OVN, Metallb, and other open-source components. Gain valuable insights into implementation challenges and the measurable performance impacts of this system designed to provide data access at scale for demanding applications. The 22-minute presentation delivers practical knowledge for managing data-intensive workloads across distributed edge environments.