SBB: Eliminating Centralized Bottlenecks in Userspace Network Runtime
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
SBB is a decentralized userspace network runtime that removes timer, monitor, and dispatcher bottlenecks by delivering user-level timer interrupts (via LAPIC) for per-request preemption and user-level NIC interrupts for CPU allocation, and by combining task stealing with flow migration for two-level load balancing. Implementation programs the NIC to migrate hot flows and uses user interrupts to wake latency-critical tasks. Evaluation shows 1.7×–5.2× higher throughput at 48 cores while meeting the same tail-latency targets and improving CPU efficiency.
AI-written summary · about
chapters (8)
- 0:00SBB user space network runtime
- 1:24Decentralized vs centralized systems
- 4:14Scheduling challenges in prior systems
- 8:50SBB’s decentralized design
- 9:30User interrupts for scheduling
- 12:43Two-level load balancing
- 14:51SBB performance evaluation
- 15:57Conclusion and future work
related talks
11mUserspace Bypass: Accelerating Syscall-intensive Applications
48mInto the hive of eBPF!
22mLIVE Q&A #3
49mBFQ I/O scheduler: more throughput, control and efficiency
16mRingLeader: Efficiently Offloading Intra-Server Orchestration to NICs
16mBolt: Sub-RTT Congestion Control for Ultra-Low Latency
50mThe Road Towards a Linux TSN Infrastructure
26mSBI Supervisor Software Events