Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-12856

Smart-NIC VF Management Port Feature

XMLWordPrintable

    • No
    • 3
    • NHE Sprint 235, NHE Sprint 236
    • 2
    • Rejected
    • False
    • Hide

      None

      Show
      None

      Description of problem:

      - Add support for Dynamic Creation Of DPU/Smart-NIC Daemon Sets and Device-Plugin Resources For OVN-K
      - DPU/Smart-NIC Daemonsets need a way to be dynamically created via specific node labels
      - The config map needs to support device plugin resources (namely SR-IOV) to be used for the management port configuration in OVN-K
      - This should enhance the performance of these flows (planned to be GA-ed in 4.14) for Smart-NIC
         5-a: Pod -> NodePort Service traffic (Pod Backend - Same Node)
         4-a: Pod -> Cluster IP Service traffic (Host Backend - Same Node)

      Version-Release number of selected component (if applicable):

      4.14.0 (Merged D/S) 
      https://github.com/openshift/ovn-kubernetes/commit/cad6ed35183a6a5b43c1550ceb8457601b53460b
      https://github.com/openshift/cluster-network-operator/commit/0bb035e57ac3fd0ef7b1a9451336bfd133fa8c1e 

      How reproducible:

      Never been supported in the past.

      Steps to Reproduce:

      Please follow the documentation on how to configure this on NVIDIA Smart-NICs in OvS HWOL mode.
       - https://issues.redhat.com/browse/NHE-550 
      
      Please also check the OVN-K daemon sets. There should be a new "smart-nic" daemon set for OVN-K.
      Please check on the nodes that the interface ovn-k8s-mp0_0 interface exists alongside ovn-k8s-mp0 interface.

      Actual results:

      Iperf3 performance:
        5-a: Pod -> NodePort Service traffic (Pod Backend - Same Node)    => ~22.5 Gbits/sec
        4-a: Pod -> Cluster IP Service traffic (Host Backend - Same Node) => ~22.5 Gbits/sec

      Expected results:

      Iperf3 performance:
       5-a: Pod -> NodePort Service traffic (Pod Backend - Same Node)    => ~29 Gbits/sec
       4-a: Pod -> Cluster IP Service traffic (Host Backend - Same Node) => ~29 Gbits/sec
      As you can see we can gain an additional 6.5 Gbits/sec performance with these service flows.

      Additional info:

      https://docs.google.com/spreadsheets/d/1LHY-Af-2kQHVwtW4aVdHnmwZLTiatiyf-ySffC8O5NM/edit#gid=88193790
      https://github.com/ovn-org/ovn-kubernetes/pull/3160

            wizhao@redhat.com William Zhao
            wizhao@redhat.com William Zhao
            Ying Wang Ying Wang
            Votes:
            0 Vote for this issue
            Watchers:
            7 Start watching this issue

              Created:
              Updated:
              Resolved: