#fdio-meeting: vpp
Meeting started by dmarion at 15:03:47 UTC
(full logs).
Meeting summary
-
- mackonstan (mackonstan,
15:04:17)
- Dave Wallace (dwallacelf,
15:04:26)
- Physical infra updates (mackonstan,
15:04:37)
- New Intel Xeon ICX based testbeds - supermicro
HW shipment (motherboard, chassis) delayed till Oct/Nov, other
components are already onsite (mackonstan,
15:05:15)
- Trending updates (mackonstan,
15:05:29)
- Daily (MRR rapid-fire tests), weekly (NDRPDR
proper perf) process: (mackonstan,
15:05:54)
- 1) nat44ed 16M sessions test, sessions not
getting established in allotted time for some tests; (mackonstan,
15:06:09)
- 2) deterministic for specific (frame_size, core
config): ipsec sw asyn tests, no traffic forwarded; (mackonstan,
15:06:19)
- Releases: CSIT-2106 report published mid-July
includes Icelake results tested externally in Intel labs using CSIT;
CSIT-2110 tracking on time (mackonstan,
15:07:08)
- CSIT-2110 release deliverables, see link
(mackonstan,
15:07:16)
- https://wiki.fd.io/view/CSIT/csit2110_plan
(mackonstan,
15:07:22)
- Virtual infra updates (mackonstan,
15:07:30)
- All FD.io systems are finally in a single DC
location hosted by Vexxhost, incl. all used LF IT systems, all
connectivity issues induced by WAN are (seemingly) behind us for
good :) (mackonstan,
15:08:05)
- session layer - support for axx socket for
application use (mackonstan,
15:09:38)
- TLS - refactored picoTLS relying on vpp crypto
infra, performance looks really good, 15 Gbps with openssl, 18 Gbps
with picoTLS, 1 connection 1 worker (mackonstan,
15:10:29)
- using sync crypto in all cases (mackonstan,
15:10:47)
- happy to discuss if testing with async crypto
makes sense (mackonstan,
15:11:07)
- can now bootstrap hoststack with binary API,
without touching VCL (mackonstan,
15:12:14)
- Damjan q: is there a need for vpp crypto infra
to support for TLS key negotiation/handling etc, apart from the data
channel supported today, and then support different crypto backends
e.g. ipsecmb (mackonstan,
15:15:22)
- Florin a: it may improve sessions/sec, will
need to check if picoTLS supports, (mackonstan,
15:17:08)
- Nathan: spent few hrs on polishing fd.io
documentation, removing old/stale info a.o. (mackonstan,
15:18:53)
- Dave Wallace: similarly, updating the FD.io
CI/infra related content; should be done before 21.10 branch is
pulled (mackonstan,
15:20:05)
- Dave W: Peter Mikus from csit converted all
CSIT docs to s3, available at usual links prefixed with "s3-"
(mackonstan,
15:20:39)
- Dave W: for vpp, optimizing the way per release
docs are generated, make test docs, doxygen, … (mackonstan,
15:21:58)
- 21.10 rc1 next week Wed 21-Sep (mackonstan,
15:22:50)
- vpp v21.01 completed, csit tests also done and
published (mackonstan,
15:23:58)
- re v21.01, v21.06 rls's - currently nothing
else planned (mackonstan,
15:24:42)
- coverity pages - still number of outstanding
defects (mackonstan,
15:25:42)
- https://github.com/vpp-dev/vpp-coverity-report
(mackonstan,
15:25:44)
- Andrew looking for feedback about the
notification mechanism to use, attempts to date are not yielding
desired effect, double digit of open defects (mackonstan,
15:26:33)
- Florin Q: asking about the process of removing
deprecated APIs (mackonstan,
15:28:40)
- Andrew A: process not fully hashed out yet, at
least 3mths(?) between marking API deprecated and removing it
(mackonstan,
15:29:14)
- https://wiki.fd.io/view/VPP/ApiChangeProcess
(dmarion,
15:32:29)
- https://wiki.fd.io/view/VPP/ApiChangeProcess
(mackonstan,
15:32:56)
- done a lot and break a lot :) (mackonstan,
15:37:24)
- big thanks to all involved (mackonstan,
15:37:31)
- deployed jcasc - jenkins configuration as
source code - system was broken for few days - but now can maintain
majority of jenkins config thru git repo in ci-managment project, so
huge improvement (mackonstan,
15:38:23)
- all HW infra now in one DC location YUL1
(mackonstan,
15:38:38)
- updated build discarder, all jenkins artefacts
deleted after 30 days, all log archives in s3 (thx again to Peter
Mikus for doing all of the integration for this) (mackonstan,
15:39:33)
- corrective note to hoststack update done by
Florin: "added support for abstract sockets for app namespaces.
Easier to use with containers" (mackonstan,
15:40:23)
- observed some git clone failures - result of
interim w/a patch to avoid WAN WireGuard links - after revert all
good again (mackonstan,
15:41:35)
- monitoring jenkins jobs daily - looking
healthier than ever (mackonstan,
15:42:14)
- Maciek areas of investment under discussion:
csit data backend, aws performance tests execution, more rack
space (mackonstan,
15:51:33)
- Matthew: what about a newer ATOM processors,
SnowRidge (mackonstan,
15:52:44)
- Damjan: problem is there are not many (none?)
commercially available systems with SNR. aimed at mobility market,
but nothing available from the open market. (mackonstan,
15:54:07)
- Maciek: putting a placeholder for this
SnowRidge/SNR ask, in addition to some new Arm/Xeon systems
(mackonstan,
15:54:52)
Meeting ended at 15:56:40 UTC
(full logs).
Action items
- (none)
People present (lines said)
- mackonstan (54)
- dmarion (3)
- collab-meetbot (3)
- dwallacelf (1)
- mortderire (1)
Generated by MeetBot 0.1.4.