#opnfv-vswitchperf: vswitchperf
Meeting started by srao at 16:03:17 UTC
(full logs).
Meeting summary
- Development Update. (srao, 16:04:01)
- Al Morton (acm,
16:04:34)
- Timestamps to Log-files- Should we?
(srao,
16:05:13)
- Adding timestamps doesn't hurt. Consolidation
of logs can be taken in future. (srao,
16:06:46)
- Testapi Container running in Pod-12.
(srao,
16:07:20)
- Add this config file: OPNFV_URL=
http://10.10.120.22:8000/a (srao,
16:07:55)
- add this into config file: OPNFV_URL=
http://10.10.120.22:8000/api/v1 (srao,
16:08:09)
- Include this when running the test: --opnfvpod
intel-pod12 (srao,
16:08:28)
- Started to explore Cedric's XTesting - create
ansible for Vsperf. Will update the team next week. (srao,
16:11:06)
- Started experimenting with newer versions of
OVS-DPDK-VPP (srao,
16:14:14)
- Need to discuss gRPC & REST limits with
Cedric. (acm,
16:14:36)
- DPDK 18.11, OVS 2.10, VPP 18.1.2 (srao,
16:15:17)
- Hari: We have to remove these flags '--burst=64
-i --txqflags=0xf00 ', if we move to DPDK 18.11 (srao,
16:18:10)
- : Support for libvirt. We have to move from
qemu-system-x86_64 command to libvirt. (srao,
16:20:38)
- Why it helps to move to libvirt: (a) For
short-duration we may not see any issues, however for longer
durations there will be performance degradation. (srao,
16:21:44)
- qemu will eventually cause a few packets to
drop with VPP (acm,
16:21:49)
- estimate of 3 weeks development time
(acm,
16:22:10)
- Why it helps to move to libvirt: (b) CPU
pinning would be relatively easier (srao,
16:23:00)
- if we move to libvirt, the XML will tell qemu
how to start, handled much more efficiently (acm,
16:23:04)
- Christian - With DPDK 18.11 We may have to
enable legacy-memory option (--legacy-mem) (srao,
16:24:02)
- DPDK 18.11 might have to add legacy memory
option, otherwise will see drops (acm,
16:24:05)
- Hari: --burst=64 i is still included, whereas
disable-hw-vlan is not (srao,
16:24:30)
- Tipsy presentation (acm, 16:29:56)
- Peter and Felician will supply slides for
posting in the read-out section of our Wiki ?? (acm,
16:57:17)
- t4p4s is the P4 switch, pronounced
"tapas" (acm,
16:58:06)
- IEEE JSAC Dec 2018 paper describes the
tool (acm,
16:58:55)
- performance results for lagopus and t4p4s were
quite low. (acm,
17:00:53)
Meeting ended at 17:03:18 UTC
(full logs).
Action items
- (none)
People present (lines said)
- srao (19)
- acm (12)
- collabot` (4)
Generated by MeetBot 0.1.4.