Chengze Liu

Notes on Distributed Storage, Consensus Protocols & Kernel Networking

Optimizing Ceph RBD Latency over 10GbE RoCE Networks

October 14, 2026 · 8 min read

When deploying distributed block storage across multi-node hyperconverged clusters, tail latency is frequently dominated by TCP congestion control and context switching overhead. Isolating replication VLANs and tuning BlueStore cache allocation brings sub-millisecond commits within reach.

Zero-Copy Packet Steering with eBPF Traffic Control

August 29, 2026 · 6 min read

Modern Linux kernels provide hooks at the XDP and TC layers that allow packet inspection and redirection before socket buffer allocation occurs.