Better Bandwidth Management with eBPF
summary
Isovalent implements eBPF-based EDT rate limiting that moves enforcement to the physical egress, uses per-packet timestamps plus a multi-queue fair qdisc timing-wheel scheduler, and avoids IFB/TC token-bucket lock contention. In a 256-concurrent-flow test (100 Mbps per flow) the EDT design shows ~7x lower median latency, ~4x lower p99 and ~7x higher transaction rate versus tc tbf. EDT needs Linux 5.1; BBR support requires ~5.18.
AI-written summary · about
chapters (10)
- 0:00Start
- 8:25The Earliest Departure Time Model
- 8:55How Can this Model Be Applied to to Kubernetes
- 9:35Sodium Bandwidth Manager
- 10:07Bpf Host Routing
- 13:20Tcp Congestion Control Algorithm
- 15:43Can Bbr Actually Be Used with Kubernetes Parts
- 16:17Kernel Limitation
- 19:36Demo
- 25:03Is pbr the Golden Solution
related talks
14mZero-overhead Container Networking with eBPF and Netkit
25meBPF Strengthens SR-IOV to Be Powerful
19mNet-P4ct: Enhanced WAN Bandwidth Fair Sharing Using P4 Programmable Switches
39mThe evolution of service mesh with eBPF
38mUnderstanding and Troubleshooting the eBPF Datapath in Cilium
34mCoupling Eventdev Usage with Traffic Metering & Policing (QoS)
22mEiffel: Efficient and Flexible Software Packet Scheduling
37mEfficient and Flexible Virtual Machine Networking Through eBPF
