DPDK and packet loss troubleshooting

[Laurent Dumont]

Hey folks,

We are troubleshooting a vendor application that is leveraging DPDK at the
application layer.

   - Openstack with RHOSP / RHEL
   - OVS as the underlying virtual switching layer. *The VM is not using
   SRIOV ports.*
   - VM acts as a load balancer (traffic pattern in IN+OUT). Traffic
   arrives on a VLAN (provider network) and exits through a VXLAN network
   (internal) to be forwarded to application servers.
   - VMs are using pinned cores
      - 4 cores total
      - 2 for the OS
      - 2 for DPDK : 1 TX one RX
   - Using hugepages
   - There are no other VMs on the physical compute.
   - compute is using isolated cores for the OS. No collision with the VM
   cores.

At low packet rates (20k~ PPS), it seems that the DPDK layer fails to
transmit packets on the TX queue. This triggers a failover.

We are not seeing anything particularly obvious on the compute side. Is
there anything at the DPDK layer that should be clear about why there is a
failure to TX?

Thanks!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/users/attachments/20220507/f5c579c5/attachment.htm>