Hi Michael, Thanks for your reply. Below are the answers to your questions inline.
On Thu, Oct 4, 2018 at 8:01 AM Michael Richardson <[email protected]> wrote: > > Onkar Pednekar <[email protected]> wrote: > > I have been experimenting with OVS DPDK on 1G interfaces. The system > > has 8 cores (hyperthreading enabled) mix of dpdk and non-dpdk capable > > ports, but the data traffic runs only on dpdk ports. > > > DPDK ports are backed by vhost user netdev and I have configured the > > system so that hugepages are enabled, CPU cores isolated with PMD > > threads allocated to them and also pinning the VCPUs. > > > When I run UDP traffic, I see ~ 1G throughput on dpdk interfaces > with < > > 1% packet loss. However, with tcp traffic, I see around 300Mbps > > What size packet? > UDP packet size = 1300 bytes TCP packet size = default packet size using iperf, the physical dpdk port MTU is 2000, dpdk vhost user port MTU is 1500, also the client running iperf alos has 1500 mtu interface, so I guess the packet size would be 1500 for TCP. What's your real pps? > UDP: dpdk and dpdkvhost user interfaces show ~ 803Kpps TCP: dpdk and dpdkvhost user interfaces show ~ 225Kpps What do you do for test traffic? > Client and server machines with iperf for generation tcp and udp traffic with 6 client threads running in parallel (-P 6 option with iperf) Client Commands: TCP: iperf -c <server ip> -P 6 -t 90 UDP: iperf -c <server ip> -u -l1300 -b 180M -P 6 -t 90 What is your latency? Are there queues full? > How can I check this? > Are you layer-2 switching or layer-3 routing, or something exotic? > OVS contains mix of l2 and l3 flows, but the (tcp/udp) traffic path uses l2 switching. > > > thoughput. I see that setting generic receive offload to off helps, > but > > still the TCP thpt is very less compared to the nic capabilities. I > > know that there will be some performance degradation for TCP as > against > > UDP but this is way below expected. > > Receive offload should only help if you are terminating the TCP flows. > I could well see that it would affect a switching situation significantly. > What are you using for TCP flow generation? Are you running real TCP > stacks with window calculations and back-off, etc? Is your offered load > actually going up? > I am using iperf to generate traffic between client and server with stable workload. > > > I don't see any packets dropped for tcp on the internal VM (virtual) > > interfaces. > > ?virtual? > I don't understand: do you have senders/receivers on the machine under > test? > By virtual i meant the dpdk vhost user interfaces. iperf is running on client and server machines external to the machine (with ovs) under test. Topology: [image: image.png] > > -- > ] Never tell me the odds! | ipv6 mesh > networks [ > ] Michael Richardson, Sandelman Software Works | network > architect [ > ] [email protected] http://www.sandelman.ca/ | ruby on > rails [ > > >
_______________________________________________ discuss mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
