#opnfv-sfc: opnfv-sfc-weekly
Meeting started by tbachman at 14:02:01 UTC
(full logs).
Meeting summary
- role call and agenda (tbachman, 14:02:24)
- tbachman (tbachman,
14:02:29)
- HELP: (DaveD_,
14:03:40)
- ebrjohn Brady Johnson (ebrjohn,
14:04:08)
- DaveD_ Dave Dolson (DaveD_,
14:04:37)
- Bryan Sullivan (bryan_att,
14:05:06)
- Paul Quinn (paulq,
14:05:10)
- (lmcdasm,
14:05:26)
- daniel Smith (lmcdasm,
14:05:31)
- repenno sent an email to the discussion list
for the demo setup (tbachman,
14:06:31)
- https://wiki.opnfv.org/service_function_chaining?&#meetings
(ebrjohn,
14:06:31)
- http://ircbot.wl.linuxfoundation.org/meetings/opnfv-sfc/2015/opnfv-sfc.2015-07-08-14.06.log.html
Minutes from last week’s meeting (tbachman,
14:06:48)
- ebrjohn says there has been a good discussion
on the list re: the orchastration/setup (tbachman,
14:07:23)
- https://docs.google.com/presentation/d/1gbhAnrTYbLCrNMhMXin0lxjyg-7IHNPjrlBTIjwAzys/edit?usp=sharing
(ebrjohn,
14:07:36)
- https://docs.google.com/presentation/d/1gbhAnrTYbLCrNMhMXin0lxjyg-7IHNPjrlBTIjwAzys/edit?usp=sharing
Presentation showing setup for SFC (tbachman,
14:08:13)
- lmcdasm says he has pictures on once br-sfc is
up, how that’s supposed to work, along with repenno’s doc on
SFC-102 (tbachman,
14:08:42)
- ebrjohn says that folks should use the ODL SFC
102 guide — it supercedes the SFC 101 guide (tbachman,
14:10:02)
- http://lists.opnfv.org/pipermail/opnfv-tech-discuss/2015-July/003733.html
email with link to SFC 102 guide (tbachman,
14:10:16)
- ebrjohn says that when ODL SFC receives
configuration for the SFs, the main piece of information in that
configuration is the service function type — firewall, QoS, DPI,
NAPT, etc. (tbachman,
14:12:34)
- ebrjohn says we also get the transport type in
the data plane locator (VXLAN, MPLS, etc.) (tbachman,
14:12:51)
- ebrjohn says the last piece of information is
the IP and port of the VTEP (tbachman,
14:13:09)
- once those 3 pieces of information are
available, that’s enough to send to the VNF Manager (tbachman,
14:13:27)
- lmcdasm says it’s a question realy of
undersanding the flow — SFC will send to the VNF and say “give me a
VM that satisifies this service type, this transport type,” ,etc. —
or is it the other way around, where the VNF asks SFC for that
information (tbachman,
14:14:18)
- lmcdasm asks which VNF manager people want to
use — tacker (sp)? He’s a bit dubious of tosca, etc. (tbachman,
14:15:03)
- bryan_att is looking into how ODL SFC etc
relates to the BGP-VPN based SFC work that we are looking into as a
Neutron-related project in OpenStack. I'd like any expert thoughts
on how the two relate. (tbachman,
14:15:14)
- bryan_att sasys he doesn’t see FC as a
higher-layer service orchestration function, rather as a mechanism
to arrange a chain out of components/resources that have already
been instantiated, e.g. known VMs and ports they have. (tbachman,
14:15:26)
- bryan_att says he doesn’t see SFC as invoking
VM instantiation on demand. That's a service orchestration
function. (tbachman,
14:15:41)
- paulq says the former is more consistent, where
the user says they need a chain of XYZ, and we can check if a chain
is already there and can be used (tbachman,
14:16:47)
- lmcdasm says he sees it the other way — the VNF
manager should be boss (tbachman,
14:17:01)
- paulq says he doesn’t think either GUI is what
the user is going to see — it will be something else (tbachman,
14:17:17)
- paulq says there would be something higher
level (tbachman,
14:17:33)
- bryan_att says there likely will be a customer
provisioning portal to do this (tbachman,
14:17:45)
- paulq says there’s a lot of other functionality
that might have to live there as well (tbachman,
14:19:59)
- Manuel Rebellon (MR_Sandvine,
14:20:01)
- paulq says at one point, over time we’re going
to want to migrate information to having constraints, more than just
the type of service (tbachman,
14:21:18)
- ebrjohn says lmcdasm can start playing with
tacker to see what’s involved (tbachman,
14:22:12)
- paulq says that to test tacker, it’s probably
easiest if there’s no service type ready (e.g. firewall) it just
goes and creates one (tbachman,
14:23:03)
- bryan_att says he still needs to look into the
SFC project details in ODL to understand the scope and focus better.
I was assuming it was a "meta-NB API" that enabled NFVO/VNFM to
connect functions/VMs as needed into a chain. But what's needed in
the chain, and whether existing or new resources are used, are
determined by the NFVO (what we call "service orchestrator").
(tbachman,
14:23:19)
- ebrjohn asks how the VNF manager knows to map
for a particular FW on a particular VM? (tbachman,
14:25:07)
- lmcdasm asks if restrictions would be the
responsibility of the coder of the portal (tbachman,
14:26:23)
- paulq says that’s consistent with what he’s
seen with cataloging systems (tbachman,
14:26:35)
- bouthors_ asks how this works with GBP
(tbachman,
14:26:48)
- paulq says GBP doesn’t know what the chain is,
per-se, but just has the action of chain (tbachman,
14:27:09)
- bouthors_ asks if the service functions are
tied to the method of service chaining implemented (e.g. this SF
can’t be in an NSH chain if it’s not NSH aware) (tbachman,
14:27:48)
- paulq says today we have not done this
(tbachman,
14:27:57)
- bouthors_ says if the SF is dependent on the
chaining methodology, then we need the capability of defining the
chain as well to ensure compatibility (tbachman,
14:28:26)
- paulq says when you hand the packet to the SF
itself, it needs NSH or a tag for descrimination (tbachman,
14:29:24)
- ebrjohn asks what’s the trigger to kick of VM
creation for SFs (tbachman,
14:31:04)
- ebrjohn says if this is from a higher level
orchestration function, what does that look like (tbachman,
14:31:21)
- lmcdasm says if we put all 3 elements in place,
we can try it a few ways. (tbachman,
14:31:47)
- lmcdasm says he imagines that ODL talking to
VNF manager might be the way to go, and probably not the other
direction (tbachman,
14:32:09)
- ebrjohn says if we do some dev on the SB REST
listener, we need to be able to tell it to go talk to the VNF
manager to create the VMs (tbachman,
14:32:53)
- paulq says the question is how do we trigger
the request out (tbachman,
14:33:11)
- ebrjohn says the trick is where to put the VNF
manager configuration (tbachman,
14:33:29)
- DaveD_ asks if we require a chain with certain
functions, if a given function already exists, which agent decides
whether the existing function is used or if a new one is spun
up (tbachman,
14:38:08)
- ebrjohn says that is part of the creation of
the chain/path/rendererd-service-path process in SFC (tbachman,
14:38:31)
- bryan_att says “it's not clear to me that ODL
is the right platform to determine whether there are
active/available resources for a chain, or whether new resources
need to be spun up. Certainly an option, but in some NFVI
environments this may be a role provided by the service
orchestrator.” (tbachman,
14:40:10)
- ebrjohn says it’s possible to do that — you can
go either way (tbachman,
14:40:26)
- bouthors_ says that when you define a rendered
service path, you have to tie a SF to an SFF (tbachman,
14:40:56)
- bryan_att says “It's certainly possible that
ODL could do this and I don't think we (as an end-user) have
developed a strong opinion or concrete alternate proposal so far...
so we need to see how well ODL's project supports this and how it
integrates/synergizes with what we have been calling our service
orchestrator (tbachman,
14:42:29)
- ebrjohn says let’s have ODL send messages to
tacker for now (tbachman,
14:43:23)
- bouthors_ asks if we could augment the SFC
model to provide the UUID of the descriptor? (tbachman,
14:44:18)
- ACTION: ebrjohn to
create sequence diagrams on the porposed architectures (tbachman,
14:47:51)
- ACTION: Paul Quinn
will setup an architecture doc (skeleton) (ebrjohn,
14:57:44)
Meeting ended at 15:09:16 UTC
(full logs).
Action items
- ebrjohn to create sequence diagrams on the porposed architectures
- Paul Quinn will setup an architecture doc (skeleton)
Action items, by person
- ebrjohn
- ebrjohn to create sequence diagrams on the porposed architectures
People present (lines said)
- tbachman (87)
- lmcdasm (16)
- ebrjohn (13)
- collabot (10)
- bryan_att (8)
- DaveD_ (5)
- MR_Sandvine (3)
- fbl (2)
- paulq (1)
- bouthors_ (1)
Generated by MeetBot 0.1.4.