#fdio-meeting: FD.io CSIT Project Meeting

Meeting started by mackonstan at 14:11:54 UTC (full logs).

Meeting summary

    1. Vratko Polak. (vrpolak, 14:12:03)
    2. (mackonstan, 14:12:05)
    3. Juraj LinkeŇ° (jlinkes, 14:12:06)
    4. Tibor Frank (tifrank, 14:12:07)
    5. Maciek Konstantynowicz (mackonstan, 14:12:15)
    6. Dave Wallace (dwallacelf, 14:12:25)
    7. Fan Zhang (fan_zhang, 14:12:26)

  1. Agenda bashing (mackonstan, 14:13:49)
  2. Inputs from projects (mackonstan, 14:13:54)
  3. Inputs from projects - VPP (mackonstan, 14:14:01)
  4. Inputs from projects - VSAP (mackonstan, 14:14:12)
  5. Inputs from projects - TSC (mackonstan, 14:14:25)
  6. New performance tests - gtpu (mackonstan, 14:15:28)
    1. Vratko, tests are merged. (mackonstan, 14:16:16)
    2. Tibor, covered in report, will add to trending. Can be removed from the agenda. (mackonstan, 14:16:53)

  7. New functional Tests - flow "offload" patch (mackonstan, 14:18:03)
    1. Juraj, comments addressed, one comment from Vratko outstanding. (mackonstan, 14:18:55)
    2. ignore two posts listed above (mackonstan, 14:20:49)

  8. New functional Tests - IPsec SPD outbound flow cache patch (mackonstan, 14:21:17)
    1. Juraj, comments addressed, one comment from Vratko outstanding. (mackonstan, 14:21:28)
    2. Vratko, 32308, review almost complete, can be merged soon. (mackonstan, 14:22:02)
    3. Vratko, 32895, waiting for VPP. (mackonstan, 14:22:14)

  9. New functional Tests - flow "offload" patch (mackonstan, 14:22:33)
    1. Fan, waiting to have the CVL fully onboarded in vpp_device testbed. (mackonstan, 14:23:52)
    2. Peter, missing QSFP28 cable for the 2nd testbed, awaiting new cable shipment (in-flight). Will be finishing the coding part later this week. (mackonstan, 14:26:47)
    3. Fan, the benchmarking is TBD. (mackonstan, 14:27:44)

  10. Release - CSIT-2106 (mackonstan, 14:28:16)
    1. Tibor, rls2106 testing going well, iterative tests, reconfig tests are almost done. (mackonstan, 14:29:06)
    2. Tibor, will be starting coverage tests (mackonstan, 14:29:22)
    3. Tibor, draft report ver in master (mackonstan, 14:29:36)
    4. Tibor, intermittent failures: 1) rdma-core on one testbed, switching driver problem; 2) create avf interface, moving thru NIC rxq configurations; 3) nat44ed 16M sessions not be established in allotted time for some tests; 4) deterministic for specific (frame_size, core config): ipsec sw asyn tests, no traffic forwarded; 5) VSAP tests with MLX (mackonstan, 14:33:56)
    5. Vratko, Peter re 5) ARP resolution issue. (mackonstan, 14:34:28)
    6. https://logs.fd.io/production/vex-yul-rot-jenkins-1/csit-vpp-perf-verify-master-3n-skx/476/ <-- Ipsec async scheduler results for 3n-skx (vrpolak, 14:34:42)
    7. https://logs.fd.io/production/vex-yul-rot-jenkins-1/csit-vpp-perf-verify-master-3n-dnv/109/ <-- for 3n-dnv (vrpolak, 14:34:58)
    8. The links are for master, but we expect 2106 to look the same. (vrpolak, 14:35:31)
    9. The performance change happened after we switched CSIT code globally to use one rxq per dataplane worker and interface. (vrpolak, 14:36:16)
    10. Vratko, multiple progressions observed. (mackonstan, 14:37:30)
    11. Vratko, 2 regressions on Xeons: 1) ipsec, 2) nat44ed minimum scale cps. (mackonstan, 14:38:40)
    12. Vratko, 1 regressions tsh: 1) vhost with vpp in vm. (mackonstan, 14:39:20)
    13. correction for above line: with testpmd in vm. (mackonstan, 14:39:40)
    14. Tibor, to complete report, the main work is static content for methodology sections per report TODO email sent last Tue (mackonstan, 14:41:14)
    15. Dave, re data comparison tables for tsh and for tx2 retest of v2101 in current CSIT env, getting failures in build jobs, working on it. (mackonstan, 14:42:30)

  11. Releases CSIT-2101 (mackonstan, 14:43:55)
    1. Dave, VPP v21.01.1 maintenance release artefacts to be produced next week (mackonstan, 14:44:40)

  12. Physical Infrastructure (mackonstan, 14:45:02)
    1. Maciek, lab extension 2021 is going ahead. (mackonstan, 14:51:04)
    2. Maciek, vendor equipment funding agreement - WIP, vendors' reps are gathering every 2 weeks to progress this, I'm coordinating. (mackonstan, 14:52:16)

  13. Virtual Infrastructure, CI/CD (mackonstan, 14:52:28)
    1. Dave, no issues (apart from fdio-yul vpn outage for EU folks) (mackonstan, 14:52:54)
    2. Juraj, new issue in vpp_Device, not all interfaces are seen in vpp after vpp start, symptom similar to previous issue we had, but root cause likely different. Detail in jira "vfio is busy when initializing VPP with multiple VPP instances using different VFs" (mackonstan, 14:55:49)
    3. https://jira.fd.io/browse/VPP-1986 (mackonstan, 14:55:56)
    4. Peter, would like to go about it differently compared to previous issue we had. Don't wait too long for resolution from Intel side, instead of switch to AVF. (mackonstan, 14:57:34)
    5. Fan, would like to see logs from Xeon platform. (mackonstan, 14:57:48)
    6. Juraj, also observed this issue with AVF. (mackonstan, 14:58:04)


Meeting ended at 14:59:24 UTC (full logs).

Action items

  1. (none)


People present (lines said)

  1. mackonstan (51)
  2. vrpolak (5)
  3. collab-meetbot (4)
  4. jlinkes (1)
  5. tifrank (1)
  6. dwallacelf (1)
  7. fan_zhang (1)


Generated by MeetBot 0.1.4.