0 views collected around this technical thread.
This article reviews ByteDance's STE team's breakthrough that reduces KVM hot‑upgrade downtime from 1000 ms to 160 ms by transparently supporting PCI passthrough devices, detailing the underlying challenges, three key kernel/hypervisor improvements, proof‑of‑concept results, and their impact on cloud IaaS operations.
By integrating OVS‑DPDK with VF‑based flow separation, stateless forwarding, and dual redundant processes, Didi Cloud’s engineering team achieved millisecond‑level hot upgrades—approximately 80 ms per VM—while maintaining high‑performance packet forwarding, scalability across hundreds of VMs, and rapid fault recovery.
This article summarizes a 2016 Global Operations conference talk that explains the challenges of cloud high‑availability, the classification of failures, and the practical solutions implemented by Kingsoft Cloud—including hot kernel upgrades, online migration, incremental disk formats, and automated backup—to keep service interruptions under a few hundred milliseconds.