Low Throughput in NSM SFC Deployment on Kubernetes
Description
I'm using Network Service Mesh (NSM) deployed on Kubernetes to chain Virtual Network Functions (VNFs) into a Service Function Chain (SFC). The setup is as illustrated in the attached architecture diagram: on Node01, there's an nse-firewall-vpp connected to passthrough-1, passthrough-2, and passthrough-3. The Client side includes cmd-nsc and iperf3, while the Gateway has nse and iperf3, with a central Forwarder handling the traffic.
To test the throughput, I ran iperf3 between the NSC and NSE. However, the observed throughput is only around 1.2 Gbps, as shown in the attached iperf3 output screenshots.
For comparison, we developed a custom forwarder based on KubeOVN, with logic similar to NSM's. In the identical environment (without DPDK or SRIOV), this achieved approximately 12 Gbps.
The pod list from my deployment is also attached for reference.
I'm wondering if the low throughput in NSM is due to a configuration or deployment issue on my end? Any insights or suggestions on optimizing this would be appreciated.
Low Throughput in NSM SFC Deployment on Kubernetes
Description
I'm using Network Service Mesh (NSM) deployed on Kubernetes to chain Virtual Network Functions (VNFs) into a Service Function Chain (SFC). The setup is as illustrated in the attached architecture diagram: on Node01, there's an
nse-firewall-vppconnected topassthrough-1,passthrough-2, andpassthrough-3. The Client side includescmd-nscandiperf3, while the Gateway hasnseandiperf3, with a centralForwarderhandling the traffic.To test the throughput, I ran
iperf3between the NSC and NSE. However, the observed throughput is only around 1.2 Gbps, as shown in the attachediperf3output screenshots.For comparison, we developed a custom forwarder based on KubeOVN, with logic similar to NSM's. In the identical environment (without DPDK or SRIOV), this achieved approximately 12 Gbps.
The pod list from my deployment is also attached for reference.
I'm wondering if the low throughput in NSM is due to a configuration or deployment issue on my end? Any insights or suggestions on optimizing this would be appreciated.