Page Comparison

Work in Progress:

1. Qemu.
2. Extracting NIC details from the deployed pod. It is straightforward for pci-id/nic-names, but get tricky with socketfiles. Socketfiles are different for memif and virtio_user.
3. Tgen+Deployment and Node-Port configuration: automation.
4. Results collection from vinperf-pod (on console results) - To decide whether the test is completed or not and collection results in PROX case.

No meeting for next 2 weeks. Next meeting  2nd March 2022.

Will be available to run tests.

• Prox:
  • VPP – Unidirectional & Bidirectional – 5 topologies
  • OVS – Unidirectional & Bidirectional – 5 topologies
• Trex:
  • VPP – 3 topologies.
    • net_memif and dummy isn’t supported.
  • OVS –
    • Current: virtio_user0
    • Error: virtio_user not supported.
• TODO: 
  • Poor performances in some scenarios*
    • Start writing the explanations- Whys?
    • Highlight issues with Traffic Generators* (Multi-core with prox)
  • Run TRex with BSwLV
  • PROX-OVS with multi-cores should work.
  • Increase test-duration to pickup transients (64)
  • Run: Bi-Di + Prox + VPP + 64Bytes (with Luc)
  • Need to come-up with ‘golden-configuration’ – What is right configuration that will give best-performance (CNI technology – vswitches).
    • separate core for each PMD
    • More cores to VPP (topology 4 and 5) – Already the max (8 cores)… so, no use..
    • To Check:
      • More cores for trex (in Pod defn. file, add resources) – Daniele to test.
      • More cores for prox is not working***
      • Increase queues (Not sure if it is possible). Memif (vpp) and vhostuser(ovs) – "Sridhar to check"
      • Swap node – configuration? (Check with Luc).