Cilium
Hard
Networking
How does Cilium's Bandwidth Manager enforce the kubernetes.io/egress-bandwidth annotation on a pod?
Select your answer:
Show answer and explanation
Answer: eBPF Earliest Departure Time pacing with the fq qdisc
Explanation
The bandwidth CNI plugin uses a token bucket filter, which scales poorly and adds latency. Cilium sets a departure timestamp on each packet in eBPF and lets the fq qdisc on the node's network device release packets at that time. The result is smoother rate limiting with less queueing delay.
More hard Cilium questions
- Which Cilium resource defines the IP ranges that Cilium hands out to Services of type LoadBalancer?
- You install Cilium with kubeProxyReplacement=true on a cluster without kube-proxy. Why do you have to set k8sServiceHost and k8sServicePort?
- In Cilium's default cluster-pool IPAM mode, what assigns each node its pod CIDR?
- What is the purpose of Cilium's egress gateway feature?
- What is the purpose of Cilium's 'prefilter' feature?
- In Cilium's tunnel mode (VXLAN or Geneve), how does the receiving node learn the security identity of a packet sent by a pod on another node?