#opnfv-meeting: Weekly Technical Discussion 148
Meeting started by bh526r at 13:00:10 UTC
(full logs).
Meeting summary
- Roll Call (bh526r, 13:00:21)
- Bin Hu (bh526r,
13:00:27)
- Fu Qiao (bh526r,
13:00:33)
- Manuel Buil (mbuil,
13:00:54)
- Mark Shostak (bh526r,
13:02:18)
- Mike Fix (bh526r,
13:02:48)
- Xu Dan (bh526r,
13:03:04)
- Yangguangzhi y00302298 (bh526r,
13:03:25)
- Trevor Cooper (bh526r,
13:04:11)
- Murtuza Khan (bh526r,
13:04:25)
- Zhang Hong Li (bh526r,
13:04:42)
- Mark Beierl (bh526r,
13:05:07)
- Continue Discussion on How to Support CNTT in OPNFV (Qiao Fu) (bh526r, 13:06:03)
- Discussion starts with "Levels of CNTT
Artefacts" by Mark Shostak (bh526r,
13:06:40)
- https://github.com/cntt-n/CNTT/blob/master/artefacts/CNTT_intro.pdf
(bh526r,
13:07:19)
- Rex Lee (bh526r,
13:07:27)
- Fu Qiao recalled Frank B (Cisco) had some
questions last week. But Frank is absent today (bh526r,
13:08:55)
- Mark Shostak briefly introduced the slides of 6
levels of CNTT artifacts (bh526r,
13:09:53)
- Each level below is more specific with more
details than level above (bh526r,
13:10:17)
- Mark Shostak described e.g. L1, L2, L3 as
examples (bh526r,
13:11:40)
- Trevor Cooper thinks it is not clear what OPNFV
needs to do, and how to map this model to OPNFV projects in order to
support this model in OPNFV (bh526r,
13:13:41)
- Mark Shostak has another diagram to
share (bh526r,
13:14:11)
- Trevor Cooper thinks it is not necessary to
solve it today, but we need to think of the work of how to make it
work in OPNFV (bh526r,
13:15:00)
- Mark Shostak thinks it is definitely a topic
CNTT will pursue (bh526r,
13:15:33)
- Mark Shostak shares another picture (which is
availabe in Github CNTT repo) (bh526r,
13:16:15)
- Fu Qiao summarized that CNTT may not know how
and what capability OPNFV can work, and OPNFV may not know how how
CNTT works, and what to expect from CNTT (bh526r,
13:18:29)
- Mark Shostak expects mutual
communication (bh526r,
13:19:13)
- Fu Qiao mentions the slides deck that Fu Qiao,
Bin Hu and Mike Fix worked out for Paris meeting, and it covers
OPNFV project structure although the work started with OVP
activity (bh526r,
13:21:29)
- Fu Qiao switched gear to present the slides of how OPNFV can be structured to support CNTT (bh526r, 13:22:23)
- Fu Qiao presented the slides of how OPNFV can
be structured to support CNTT (bh526r,
13:22:57)
- Trevor Cooper asked how Ref M and Ref A can be
integrated into OPNFV releases (bh526r,
13:33:13)
- One Ref A can a feature project in OPNFV, which
can be implemented as one scenario by multiple installers
(bh526r,
13:34:21)
- Trevor Cooper mentioned hardware profiles, if
there are several h/w profiles, when in implementation, we need to
pick one, e.g. compute profile, and interfaces and extensions, then
we have to make decision in implementation, then set up lab
etc. (bh526r,
13:36:34)
- So not sure if it is feasible to have one
profile that is enough, or we need variations etc. (bh526r,
13:37:11)
- Then who makes decision, TSC or projects
etc (bh526r,
13:37:29)
- Mark Shostak says that CNTT will give guidance
regarding profile and variations. (bh526r,
13:37:57)
- Trevor Cooper thinks it important to have
guidance from CNTT (bh526r,
13:38:29)
- Fu Qiao thinks it is exact reason to have
feature project in OPNFV and communicate (bh526r,
13:39:17)
- Trevor Cooper has one more question regarding
installers. Now there is one installer project targeted for CNTT
implementation, e.g. Airship (bh526r,
13:40:38)
- Fu Qiao answered that it is not
restrictive (bh526r,
13:42:50)
- Bin mentioned that ideally, we encourage all
installers to implement Ref A. Of course, if installer is willing to
implement it or not is another question. But we encourage all
installers to implement all Ref A. (bh526r,
13:44:15)
- Fu Qiao mentioned that if one installer is not
willing to implement CNTT, then TSC should decide if OPNFV still
needs to host that installer that does not support CNTT (bh526r,
13:45:19)
- Manuel asked if Ref A has 3 profiles
(bh526r,
13:46:03)
- Bin mentioned Ref M has 3 profiles. Ref A work
just started. Mark S confirmed it is correct (bh526r,
13:46:36)
- Trevor gives an example: for h/w profile,
compute-intensive and network-intensive specify different RAM
requirement (bh526r,
13:53:48)
- profile specifies different RAM size as h/w
requirement (bh526r,
13:54:10)
- Then we need 2 different implementation
(bh526r,
13:54:22)
- Bin mentioned that it is only one
implementation for this specific example. When installer starts to
deploy the NFVi based on Ref A, the very 1st step is to check h/w
capability, e.g. RAM size, vCPU etc. (bh526r,
14:01:04)
- For compute-intensive, it checks if RAM size
meets the minimum RAM requirement. The same for
network-intensive. (bh526r,
14:01:41)
- If h/w capability meets the minimum
requirement, then move on to continue rest steps of
deployment (bh526r,
14:02:08)
- If h/w capability does not meet the minimum
requirement, then reports an error: Insufficient RAM or vCPU or
something else (bh526r,
14:02:45)
- So in this case, there is only one
implementation, and the implement checks h/w capability as the 1st
step in deployment (bh526r,
14:03:17)
- Those h/w profile are in configuration, nothing
is hard-coded. So the implementation is scalable (bh526r,
14:03:57)
- Bin suggested that we continue to discuss it
next week. (bh526r,
14:04:27)
- Trevor Cooper and Mark Shostak may prepare more
examples of variations, and we can further discuss whether or not
those mean different implementations or actually one implementation
to support multiple variations in runtime (bh526r,
14:05:21)
- AOB (bh526r, 14:05:30)
- We continue to discuss next week with more
examples from Trevor C and Mark S (bh526r,
14:06:02)
- Meeting adjourned (bh526r,
14:06:08)
Meeting ended at 14:06:12 UTC
(full logs).
Action items
- (none)
People present (lines said)
- bh526r (59)
- collabot` (3)
- mbuil (1)
Generated by MeetBot 0.1.4.