Hi, I am experiencing very poor throughput performance on traffic flowing out of my VPP VM. I had initially thought it was due to the tap-cli interfaces which were relaying traffic from a kernel interface to VPP and performing CG-NAT along its way to the outside network. However, after implementing a simple test, it seems like an inherent issue of VPP performance on virtual machine.
To experiment a basic topology, I followed Intel's guide to VPP mentioned in this article ( https://software.intel.com/en-us/articles/build-a-fast-network-stack-with-vpp-on-an-intel-architecture-server ) which is setup as: Server 1 [iperf3 client] <-10G-> Server 2 [VPP] <-10G-> Server 3 In the article, they are using 40G links between the servers, but I am limited in hardware to 10G NICs. My 3 servers are Dell R620 with dual socket Xeon E5-2650 8 core processors with 64GB RAM using 82599ES 10-Gigabit SFI/SFP+ NIC cards. I am using Ubuntu 16.04 server OS. I followed the guide. Using kernel routing/forwarding, I achieved 9.2 Gbps on iperf and after binding interfaces to VPP on server 2, I also obtained 9.2-9.3 Gbps throughput on iperf3 using TCP as per above article. The issue comes when you replicate the same setup in form of KVM VMs using virt-manager/libvirt. Here is my VM topology on server 2 (Ubuntu 16.04 server on both host OS as well as guest OS): VM 1 [iperf client] <-virtio interface-> VM 2 [VPP] <-virtio interface-> VM 3 [iperf server] Replicating the same test, using kernel routing I am getting 16 Gbps throughput using iperf3. After binding interface to VPP however, I achieve just 1.3 Gbps. After following optimization guide in the above Intel article (hugepages in 80-vpp.conf, main core, corelist-workers in startup.conf), speed was still 1.3-1.4 Gbps. Later, I also applied KVM optimizations as per VPP Wiki ( https://wiki.fd.io/view/VPP/How_To_Optimize_Performance_(System_Tuning) ) but I am still getting no improvements at all. In summary, Without VPP: 16 Gbps throughput With VPP: 1.3 Gbps throughput Has anyone else encountered this much drop in performance and any workarounds to fix this. Regards, Hamid -=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#9571): https://lists.fd.io/g/vpp-dev/message/9571 Mute This Topic: https://lists.fd.io/mt/21793216/21656 Mute #vpp: https://lists.fd.io/mk?hashtag=vpp&subid=1480452 Group Owner: vpp-dev+ow...@lists.fd.io Leave: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
