#opendaylight-group-policy: gbp_status_arch
Meeting started by tbachman at 18:00:11 UTC
(full logs).
Meeting summary
- agenda (tbachman, 18:00:18)
- https://wiki.opendaylight.org/view/Group_Policy:Sub-Groups:STATUS#Team_Meeting
Agenda for today’s meeting (tbachman,
18:00:31)
- https://meetings.opendaylight.org/opendaylight-group-policy/2015/gbp_150206/opendaylight-group-policy-gbp_150206.2015-02-06-17.58.html
Minutes from last week’s meeting (tbachman,
18:00:42)
- https://docs.google.com/document/d/1ZnLHuSiyOmVFR0pW65K4enB7QcYbLL18yzxYgmK_iDo/edit
Google document that alagalah created, as an action from last week’s
meeting on SFC/GBP integration (tbachman,
18:01:33)
- Code merges (alagalah, 18:02:55)
- alagalah says there was a bug fix merged by
Intel (tbachman,
18:06:12)
- alagalah says we’re hoping to merge the
opencontrail renderer patch — just need to run the POC to make sure
there’s no change functionally (tbachman,
18:06:42)
- Neutron impl update (alagalah, 18:07:37)
- alagalah met with martin_sunal, who will be
helping with the neutron work (tbachman,
18:09:02)
- alagalah says he’s kitted up a bulk of the work
for multiple EPGs per EP, and intra-group policy, and tested
it (tbachman,
18:09:35)
- alagalah says from a control plane, it works
great (tbachman,
18:09:42)
- alagalah says the data plane — specifically the
flow pipeline — may need to be worked out a bit, with multiple
EPGs (tbachman,
18:10:07)
- alagalah is looking at the issues with
multi-EPGs (tbachman,
18:10:21)
- alagalah is going to compare doing this vs.
trying to do this using condition groups — provide a comparison
between the two (tbachman,
18:10:48)
- alagalah says martin_sunal had the idea of an
“ANY” EP — to make a bit of a “grey list” model (instead of white
list) (tbachman,
18:12:35)
- alagalah says he’s looking at the merits of all
of these changes (tbachman,
18:12:54)
- SFC and GBP Integration (continued) (alagalah, 18:13:01)
- https://docs.google.com/document/d/1ZnLHuSiyOmVFR0pW65K4enB7QcYbLL18yzxYgmK_iDo/edit
Google document that alagalah created, as an action from last week’s
meeting on SFC/GBP integration (tbachman,
18:13:16)
- alagalah says GBP and SFC are working to
integrate for an external application — working on a POC for
Lithium (tbachman,
18:13:54)
- https://wiki.opendaylight.org/view/Service_Function_Chaining:Group_Based_Policy_Integration
Wiki page that repenno created for this integration (tbachman,
18:14:21)
- mickey_spiegel asks if we’re relying on
separation of topology, encaps or both for conflict avoidance
(tbachman,
18:15:51)
- alagalah says we may require a constrainted
topology, and will require NSH (tbachman,
18:16:06)
- ebrjohn says he’d like to see at a minimum that
the first solution be flexible enough to support different encaps —
specifically MPLS (tbachman,
18:17:01)
- alagalah asks where that would sit in the
arch (tbachman,
18:17:36)
- ebrjohn says to use MPLS instead of NSH
(tbachman,
18:17:51)
- ebrjohn says NSH doesn’t have many switches
supporting it (tbachman,
18:18:04)
- ebrjohn says there’s an NSH proxy, which sits
between the switch and the service function (tbachman,
18:18:25)
- ebrjohn is looking to support MPLS since it’s
more common (tbachman,
18:18:38)
- alagalah says he thinks it’s achievable, but is
still interested in NSH (tbachman,
18:19:06)
- ebrjohn says he isn’t advocating excluding NSH,
but rather to do both — matter of resources and time (tbachman,
18:19:35)
- ebrjohn says he can do it in SFC (tbachman,
18:19:41)
- ebrjohn asks if GBP supports MPLS (tbachman,
18:19:57)
- alagalah says no, but there’s also currently no
NSH encap either yet (tbachman,
18:20:09)
- alagalah says if we can get an MPLS encap from
SFC, then we should be able to make something work (tbachman,
18:20:30)
- ebrjohn says he’s willing to help contribute
that code in the GBP project (tbachman,
18:20:41)
- Prem says that for MPLS, the label would be
used to create a similar path as NSH; asks what the benefit is of
both NSH and MPLS (tbachman,
18:21:14)
- ebrjohn says he wasn’t thinking both at the
same time — just one or the other (tbachman,
18:21:29)
- paulq says there’s a case that might bear
mentioning, where MPLS and NSH are used; allows extending policy to
the services, but use MPLS for the transport (tbachman,
18:24:15)
- ebrjohn says that in the case where there are
networks that don’t support NSH, the MPLS use case is helpful
(tbachman,
18:24:48)
- ebrjohn says you can also do MPLS over
GRE (tbachman,
18:25:51)
- paulq says it’s important to consider one goal
of service chaining is the service, and extending policy to the
service can be good (tbachman,
18:26:49)
- rukhsana asks about the acceptance of the NSH
patches in OVS (tbachman,
18:27:17)
- paulq says the patches are in, work, and work
is in place to get them upstreamed (tbachman,
18:27:36)
- abhijitkumbhare asks what the timeline is for
phase 0 (tbachman,
18:29:39)
- alagalah says Lithium (tbachman,
18:31:07)
- abhijitkumbhare asks mickey_spiegel what no
encap means — no NSH? (tbachman,
18:31:38)
- mickey_spiegel asks if VxLAN stitching falls
under no-encap (tbachman,
18:32:00)
- abhijitkumbhare says no encap is
reclassification at every hop (tbachman,
18:32:29)
- paulq says VxLAN is encap (tbachman,
18:32:59)
- paulq says SFC needs to hand back what can be
used in the chain (tbachman,
18:35:17)
- paulq says that’s the only way to make this
generalized (tbachman,
18:35:42)
- mickey_spiegel asks about the stuff that
readams was bringing up on the mailing list (tbachman,
18:35:47)
- https://lists.opendaylight.org/pipermail/groupbasedpolicy-dev/2015-February/000903.html
email from Rob Adams, talking about mulitiple renderer issue (tbachman,
18:36:36)
- alagalah recommends addressing conflict
detection/resolution in phase 1 (tbachman,
18:38:02)
- ebrjohn says we may want to consider the
overlay API as a longer-term goal (tbachman,
18:39:21)
- mickey_spiegel asks when SFC says Rendered
Service Path, does that mean selecting what the SFFs are, or
creating the data path things as well (tbachman,
18:41:48)
- paulq says it’s the actual toplogy created for
hte service path — SFF1, switch62, FW3, etc. (tbachman,
18:42:09)
- repenno says a service function path is a
collection of service hops that a user can input as a constraint for
constructing a rendered service path (tbachman,
18:44:37)
- mickey_spiegel asks if it includes the encap
details between them (tbachman,
18:45:47)
- repenno says it includes everything if the data
plane locators and encaps were provided; RSP provides an ordererd
list of SFFs and SFs; the data plane locators are sent through
another interface/API (tbachman,
18:46:42)
- repenno then looks up the data plane locators
and encaps when constructing the RSP (tbachman,
18:47:04)
- repenno says if folks want to understand SFC,
he gave a demo at the SFC meeting which should help (tbachman,
18:48:54)
- https://meetings.webex.com/collabs/meetings/playRecording?recordID=14433636&meetingInstanceID=I3RF8SUSATNXF89CT7CW0PWB30-9VIB
Link to SFC presentation from repenno on SFC (tbachman,
18:49:59)
- https://wiki.opendaylight.org/view/Service_Function_Chaining:Main#Meetings
(ebrjohn,
18:50:05)
- https://meetings.webex.com/collabs/url/VvVrUGAbeDLYPAxlPgmw5TlduQHGLb1gjk5_E1rxDum00000
Alternate link to webex for non-US accounts (tbachman,
18:50:22)
- alagalah says openstack GBP already integrates
with ODL GBP today (tbachman,
18:50:36)
- alagalah says openstack SFC, to his knowledge,
has a FW and LB encapsulated in it — when they ask for an SFC,
they’re just leveraging the router’s capabilities, and is relatively
static (tbachman,
18:51:09)
- alagalah asks if the SFC folks are looking to
create a neutron API implementation, or is that out of scope
(tbachman,
18:51:44)
- repenno says he looked into that, and is on his
todo list (tbachman,
18:52:02)
- alagalah says the GBP neutron provider would
also look to support SFC from openstack (tbachman,
18:52:31)
- repenno asks if GBP is doing the neutron stuff
— is it possible to put this stuff outside GBP or put it in
GBP (tbachman,
18:52:53)
- alagalah says GBP will create a provider for
neutron, so it can integrate with existing neutron (tbachman,
18:53:26)
- s3wong says the current Service Chaining
implementation is part of the openstack GBP repo (tbachman,
18:54:03)
- ebrjohn asks if we can get GBP and SFC
integrated and it works well, do we need to do this in ODL?
(tbachman,
18:54:37)
- alagalah says he doesn’t want to preclude SFC
from doing any integration that their project is interested in
doing (tbachman,
18:55:00)
- repenno says we need to wait and see what that
looks like (tbachman,
18:55:09)
- songole says openstack has a firewalll and load
balancer today; (tbachman,
18:57:17)
- s3wong says Neutron has no "service insertion"
framework, the reference doesn't really have a good data path
plumbing (tbachman,
18:58:24)
- s3wong says there is a service chain API/plugin
infra on the OpenStack GBP repo (tbachman,
18:58:51)
- s3wong says the point is, since we can’t
translate: map a flow match a classifier to direct that traffic flow
to a service, the data path part would be great to have an
Open-source reference that truly works (tbachman,
18:59:46)
- https://docs.google.com/presentation/d/1AO47EYrDMuTAypcpYbL3XOKFggZXu9x00XpGVwJLr9o/edit#slide=id.g612a6349b_0205
slides for neutron integration (tbachman,
19:00:37)
- alagalah asks what the best way is to track
inter-project work? Trello? (tbachman,
19:03:01)
- ebrjohn says he’s fine (as SFC PTL) for hosting
this on the GBP Trello (tbachman,
19:03:14)
- https://trello.com/b/yc0xHFlv/opendaylight-groupbasedpolicy-lithium
GBP Trello Board, with GBP/SFC integration cards (tbachman,
19:03:32)
- https://git.opendaylight.org/gerrit/#/c/15255/
initial patch to start SFC/GBP conversation (tbachman,
19:07:17)
Meeting ended at 19:08:41 UTC
(full logs).
Action items
- (none)
People present (lines said)
- tbachman (119)
- s3wong (12)
- odl_meetbot (8)
- ebrjohn (5)
- alagalah (3)
- abhijitkumbhare (2)
- gzhao (1)
- dbainbri1 (1)
Generated by MeetBot 0.1.4.