BestHub
Sign in
Tencent Technical Engineering
Tencent Technical Engineering
Feb 11, 2026 · Backend Development

How WFS Ultra Achieved 200 Gbps TCP Throughput, Surpassing RDMA‑Based 3FS

The article details how the WFS Ultra project re‑engineered a traditional TCP‑based distributed file system with a Run‑To‑Completion thread model, ultra‑core async networking, full‑link zero‑copy, and load‑adaptive prefetch, enabling 200 Gbps Fio throughput that exceeds the RDMA‑accelerated 3FS benchmark.

Run-to-CompletionTCP performanceWFS Ultra
0 likes · 10 min read
How WFS Ultra Achieved 200 Gbps TCP Throughput, Surpassing RDMA‑Based 3FS
Liangxu Linux
Liangxu Linux
Dec 9, 2024 · Fundamentals

Master Linux Zero‑Copy: Sendfile, Splice, mmap+write, and tee Explained

This article explains how Linux zero‑copy techniques—DMA, sendfile, splice, mmap + write, and tee—reduce CPU involvement in large file and network transfers by moving data directly within kernel space, detailing their workflows, code examples, performance trade‑offs, and suitable use cases.

DMALinuxSystem Programming
0 likes · 20 min read
Master Linux Zero‑Copy: Sendfile, Splice, mmap+write, and tee Explained
JavaEdge
JavaEdge
Feb 26, 2023 · Backend Development

Which Network I/O Model Should You Choose for RPC Design?

This article explains the role of network communication in RPC, compares blocking I/O, non‑blocking I/O, I/O multiplexing and asynchronous I/O, discusses why blocking I/O and multiplexing dominate, introduces zero‑copy techniques at the OS and Netty levels, and shows how Netty optimizes data handling for high‑performance RPC.

NettyNetwork I/ORPC
0 likes · 11 min read
Which Network I/O Model Should You Choose for RPC Design?