#fdio-vpp: fdio-vpp

Meeting started by DaveBarach at 15:09:08 UTC (full logs).

Meeting summary

  1. CSIT (Maciek reporting) (DaveBarach, 15:10:50)
    1. New tests: (mackonstan, 15:11:23)
    2. still onboarding new tests, including nat44 tests (DaveBarach, 15:11:56)
    3. Lots of nat44: i) nat44det throughput (stateless, bi-dir), ii) nat44ed throughput (stateless, uni-dir), iii) nat44ed CPS and PPS throughput (stateful, bi-dir). (mackonstan, 15:12:03)
    4. All nat44 tests at varying scale from 64k to 16mil UDP (stateless, stateful) and TCP (stateful) sessions. (mackonstan, 15:12:11)
    5. TRex upgraded to v2.82, calibration work for STL and ASTF. (mackonstan, 15:12:35)
    6. AMD 2n-zen2 EPYC 7532 testbed onboarded, wip dry run tests, CSIT-2009 tests to start shortly. (mackonstan, 15:13:15)
    7. CSIT-2009 bulk of VPP v20.09 tests done, initial report with results to be published tomorrow, wip draft report at the link below (mackonstan, 15:13:31)
    8. https://docs.fd.io/csit/master/report/ (mackonstan, 15:13:38)
    9. https://docs.fd.io/csit/master/report/ (mackonstan, 15:13:48)
    10. Report to be updated next week and week after with more runs/tests incl. AMD and nat44ed stateful. (mackonstan, 15:14:02)
    11. Maciek to communicate location of AMD epyc tuning notes (DaveBarach, 15:15:31)

  2. Host Stack (Florin) (DaveBarach, 15:15:41)
    1. session layer added pure socket-based APIs (DaveBarach, 15:15:56)
    2. will deprecate binary API attachment interface (DaveBarach, 15:16:09)
    3. cleaner / more secure way of integrating apps, particularly in containers (DaveBarach, 15:16:26)
    4. TCP small timer wheel refactor, to support 100us timer-wheel resolution (DaveBarach, 15:16:56)
    5. to support more advanced RTT-level timing (DaveBarach, 15:17:35)
    6. Envoy integration upstream refactoring, to facilitate host-stack / Envoy comms (DaveBarach, 15:18:14)
    7. overview paper will be presented at EnvoyCon in two days (DaveBarach, 15:18:42)
    8. florin to share recording, slides etc. (DaveBarach, 15:19:23)

  3. Documentation (DaveBarach, 15:20:13)
    1. need to find an owner: John DeNisco has retired (DaveBarach, 15:20:29)

  4. Release Management (Andrew / Dave W) (DaveBarach, 15:20:40)
    1. 20.09 update: increased "calm-down" interval definitely helped (DaveBarach, 15:21:18)
    2. Proposed plan for 21.01 dates on wiki (DaveBarach, 15:21:48)
    3. F0 on 12/09, formal release on 1/27 (DaveBarach, 15:22:38)
    4. no comments on the posted schedule, so we will go with them... (DaveBarach, 15:24:51)
    5. 19.08 deprecation discussion, should turn off the jobs around the first of the year (DaveBarach, 15:26:26)
    6. Shoot for about one quarter's worth of LTS release overlap (DaveBarach, 15:26:55)

  5. Infra (Dave W) (DaveBarach, 15:27:12)
    1. a couple of additional Jenkins outages (DaveBarach, 15:27:24)
    2. pushing the LF to root-cause the outages (DaveBarach, 15:27:42)
    3. link to EnvoyCon2020 presentation https://wiki.fd.io/images/0/08/Using_vpp_as_envoys_network_stack.pdf (florinc, 15:28:19)
    4. Might want to stop tracking "rechecks" (DaveBarach, 15:28:24)
    5. adding verify jobs for Debian 9, Debian 10 (DaveBarach, 15:31:06)
    6. Nomad infra was manually managed, transitioned to management by Ansible (DaveBarach, 15:31:44)
    7. jobs running on Ubuntu 18.04 (DaveBarach, 15:32:06)
    8. generation of Docker containers automated (DaveBarach, 15:32:36)
    9. keep Centos 7 jobs for LTS release(s) only, deprecate for master (DaveBarach, 15:40:31)

  6. Static Analysis (DaveBarach, 15:48:39)
    1. 17 outstanding, please clean up (DaveBarach, 15:48:47)

  7. Main heap / bihash-related changes (DaveBarach, 15:49:09)
    1. bihash using main heap, allows folks to use hugepages (DaveBarach, 15:49:24)
    2. main heap usage increased, but private mmap areas gone, and gone from core dumps (DaveBarach, 15:49:59)
    3. ip4 mtrie, same story (DaveBarach, 15:50:17)
    4. startup config: can specify main heap size, main heap page size (DaveBarach, 15:50:50)
    5. hugepage backing store / lack of TLB miss walks significantly improves forwarding perf w/ big tables (DaveBarach, 15:51:56)

  8. long compilation issues (DaveBarach, 15:53:43)
    1. lots of inlining. Most of the time for a good reason, but we need to avoid making the compiler go crazy (DaveBarach, 15:54:31)
    2. amd 2n-zn2 BIOS settings (mackonstan, 16:00:11)
    3. https://gerrit.fd.io/r/gitweb?p=csit.git;f=docs/lab/testbeds_sm_zn2_hw_bios_cfg.md;hb=refs/changes/33/29033/2#l483 (mackonstan, 16:00:11)
    4. zen2 servers firmware inventory (mackonstan, 16:00:11)
    5. https://gerrit.fd.io/r/gitweb?p=csit.git;f=docs/lab/testbeds_sm_zn2_hw_bios_cfg.md;hb=refs/changes/33/29033/2#l627 (mackonstan, 16:00:11)
    6. amd zen2 GRUB cmd line: (mackonstan, 16:00:11)
    7. GRUB_CMDLINE_LINUX="isolcpus=0-15,16-31,32-47,48-63 rcu_nocbs=0-15,16-31,32-47,48-63 transparent_hugepages=never iommu=pt amd_iommu=on nohz=off numa_balancing=disable nmi_watchdog=0 audit=0 nosoftlockup processor.max_cstate=0" (mackonstan, 16:00:11)


Meeting ended at 16:00:44 UTC (full logs).

Action items

  1. (none)


People present (lines said)

  1. DaveBarach (43)
  2. mackonstan (16)
  3. collab-meetbot (3)
  4. florinc (1)


Generated by MeetBot 0.1.4.