Lightning Talk: Optimizing Cluster Workloads: Cilium Envoy on DPU
summary
A proof of concept deploys Envoy as a DaemonSet on an OCTEON10 DPU and implements a cni offload layer plus plugins to present pods with virtual function pairs into an eBPF datapath, aiming for full network offload. Profiling shows host usage reaching 35% Cilium / 42% Envoy under stress, motivating the move. The architecture claims to free host CPU, use DPU hardware accelerators for L4/L7 work, and reduce cluster power by shifting common infrastructure workloads onto DPUs.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:23Proposal
- 1:32Model
- 2:35Offloading
- 2:52Full Primary Network offload
- 4:17Transition to DPU
related talks
13mLightning Talk: Accelerating Cilium: Our Journey on Offloading Cilium To DPU
10mLightning Talk: Envoy @ Netflix
19mCluster Optimization: Harnessing DPUs for Ambient Mesh Workloads
10mLightning Talk: Closing the Gap: Fair North–South Bandwidth Management for Tenants
38mEnvoy Deep Dive
10mLightning Talk: Connecting Klusters on the Edge with Deep Dive into Cilium Cluster Mesh
21mLightning Talk: Integrating Envoy with DPDK-based virtual networks
26mIntegrating Stateful Load Balancing At Scale In Envoy