From Classical to the Future
summary
Project builds a new Crimson OSD rewrite using Seastar futures/promises to cut kernel context switches into cheap user-space task switches, targeting per-core CPU overhead that prevents exploiting ~500k IOPS NVMe drives. Presenter measures HDD vs NVMe IOPS and claims a per‑IO CPU budget drop (roughly 20M cycles for spinners vs ~6k cycles for NVMe), implements single‑thread‑per‑core object contexts and message passing, ships basic RADOS/RBD functionality, and reports double‑digit throughput gains plus much lower tail latency.
AI-written summary · about