Rakaia: Scalable In-Kernel Scheduling for TCP-Based RPCs
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
Builds a message-oriented API inside the Linux kernel by parsing TCP receive streams early, implemented as a 3000-line kernel module plus a 60-line registration patch and kTLS support. Introduces a logically centralized FIFO queue realized with best-of-two placement to avoid kernel contention and a TCP delegation algorithm to enqueue messages to the same connection without taking tcp_lock. Evaluation shows up to 5× higher throughput-under-SLO vs KCM, gRPC-Go up to 1.56× and gRPC-C++ up to 2.69×, and real-app gains (Silo TPC-C 1.39×, OpenTelemetry 1.42×).
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:40gRPC popularity reasons
- 2:48Head-of-line blocking in gRPC
- 4:58Research question and Rakaya
- 6:44Rakaya's message scheduling
- 8:11Rakaya implementation
- 8:55Evaluation setup
- 9:39Synthetic RPC performance
- 10:21gRPC vs Rakaya performance
- 11:51Conclusion and open-source
related talks
20mKRCORE: A Microsecond-scale RDMA Control Plane for Elastic Computing
31mA Practical, Research‑Backed Introduction To Cloud‑Native Kernel‑Bypass Networking
21mTCP ≈ RDMA: CPU efficient Remote Storage Access with i10
01h 21mTutorial: Developing High-performance Networking Applications
17mRearchitecting the TCP Stack for I/O-Offloaded Content Delivery
21mR2P2: Making RPCs first-class datacenter citizens
24mPREEMPT_RT Q&A with tglx
17mOpening Up Kernel-Bypass TCP Stacks