Canvas - Isolated and Adaptive Swapping for Multi-Applications on Remote Memory

Explore a 14-minute conference talk from USENIX NSDI '23 that introduces Canvas, an innovative swap system designed for multi-application remote memory environments. Delve into the challenges of interference and unexpected slowdowns in existing remote memory techniques when multiple applications run simultaneously. Discover how Canvas achieves full isolation of swap paths for remote-memory applications, allowing each to have its own dedicated swap partition, swap cache, prefetcher, and RDMA bandwidth. Learn about three adaptive optimization techniques developed for Canvas: adaptive swap entry allocation, semantics-aware prefetching, and two-dimensional RDMA scheduling. Gain insights into how these features minimize performance variation and significantly reduce performance degradation across widely-deployed applications.