#opendaylight-sfc: sfc_weekly
Meeting started by tbachman at 17:00:00 UTC
(full logs).
Meeting summary
- agenda (tbachman, 17:00:13)
- https://meetings.opendaylight.org/opendaylight-sfc/2015/sfc_weekly/opendaylight-sfc-sfc_weekly.2015-02-05-17.00.html
Last week’s meeting minutes (tbachman,
17:00:23)
- https://lists.opendaylight.org/pipermail/sfc-dev/2015-February/000907.html
Agenda for today’s meeting (tbachman,
17:01:20)
- SFC/GBP integration (tbachman, 17:06:09)
- https://wiki.opendaylight.org/view/Service_Function_Chaining:Group_Based_Policy_Integration
wiki page describing the integration (tbachman,
17:08:48)
- ebrjohn says the EPGs determine which traffic
gets sent into the chain (tbachman,
17:09:07)
- SFC is responsible for getting the packets
through the chain and to the termination point (tbachman,
17:09:27)
- alagalah says it would help to frame the
conversation around scope (tbachman,
17:10:43)
- alagalah asks ebrjohn if it’s okay to delineate
scope (tbachman,
17:11:14)
- ebrjohn says its good to make clear that it’s a
POC (tbachman,
17:11:31)
- paulq agrees with addressing scope (tbachman,
17:12:01)
- alagalah says someone else provisions the
service chain (tbachman,
17:12:37)
- alagalah says there are two EPGs, with
instances on two different servers, can talk over VxLAN
tunnel (tbachman,
17:12:59)
- alagalah says one of the things between the two
EPGs is the contract with an action called CHAIN, with an action
referenced by name (tbachman,
17:13:19)
- alagalah says there’s an outer list and an
inner list — the inner list says how to get in and how to come
out (tbachman,
17:13:40)
- in the reverse direction is also needed — the
entry and exit point (tbachman,
17:13:55)
- GBP has to then encap in NSH and get it there —
black boxing SFC (tbachman,
17:14:10)
- paulq says that sounds fine (tbachman,
17:14:23)
- abhijitkumbhare asks for the prototype — will
it be possible to extend it to other transport types such as
MPLS (tbachman,
17:14:59)
- alagalah says it’s a bit out of scope of a POC
goal, as it’s specific on GBP — maybe address this in the GBP
meeting (tbachman,
17:16:18)
- ebrjohn says the communication between GBP and
SFC is NSH — does it always have to be NSH (tbachman,
17:16:44)
- alagalah says this is just to get this to the
head of the SFC chain — how we get it from the beginning or end of
the service chain to the beginning or ending endpoint is a GBP
specific thing, and maybe best to address it in that team
meeting (tbachman,
17:17:43)
- paulq says we’re assuming that GBP is receiving
the requisite information to start the chain — be it NSH, et.
al. (tbachman,
17:18:12)
- ebrjohn says it needs to be flexible enough to
use SFC with other technologies than NSH (tbachman,
17:19:06)
- paulq emphasizes that SFC will return the
requisite information for the appropriate encap (tbachman,
17:19:32)
- mikey_spiegel says that the requirement of
transport of SFC vs. transport of what GBP normally does, and as
long as they’re different, it’s okay (tbachman,
17:20:07)
- paulq doesn’t agreee with that statement
(tbachman,
17:20:14)
- alagalah says he kind of agrees with it, but
has a question on that (tbachman,
17:20:26)
- alagalah says with NSH there’s no chance of
openflow “glare” with GBP (tbachman,
17:20:50)
- paulq says he may have misunderstood
mickey_spiegel’s comment (tbachman,
17:21:17)
- mickey_spiegel says it’s only the multiple
writers that he’s addressing (tbachman,
17:22:17)
- paulq agrees (tbachman,
17:22:21)
- alagalah recommends a phased approach for long
term (tbachman,
17:23:31)
- ebrjohn agrees — what would be the next
step (tbachman,
17:24:20)
- https://trello.com/b/yc0xHFlv/opendaylight-groupbasedpolicy-lithium
Group Based Policy Trello board with cards for SFC (tbachman,
17:25:12)
- vina_ermagan asks when you say terminating EPG,
is that the final SFF, or the where the final endpoint is
located (tbachman,
17:26:06)
- ebrjohn says it’s after the SFF (tbachman,
17:26:13)
- vina_ermagan says that information has to be
conveyed to SFC so that it knows where to send it (tbachman,
17:26:24)
- GBP would ask for the chain and at the last
hop, send it to this location (tbachman,
17:26:37)
- ebrjohn says the difference between SFC as it
is now vs. sending it to the terminating EPG — is there any
difference for SFC? (tbachman,
17:27:38)
- paulq says there shouldn’t be — as long as we
can impose the classification (tbachman,
17:27:48)
- https://docs.google.com/document/d/1ZnLHuSiyOmVFR0pW65K4enB7QcYbLL18yzxYgmK_iDo/edit
(alagalah,
17:27:59)
- mickey_spiegel says there’s the issue of
creating the tunnel from the SFF (tbachman,
17:28:02)
- https://docs.google.com/document/d/1ZnLHuSiyOmVFR0pW65K4enB7QcYbLL18yzxYgmK_iDo/edit
SFC+GBP phased design brainstorming (alagalah,
17:28:13)
- mickey_spiegel says there’s an action at the
end of the chain that’s GBP-ish for the tunnel (tbachman,
17:28:41)
- vina_ermagan says that the way the POC is being
designed, it’s kind of boxing out SFC — control of everything,
including the last SFF (tbachman,
17:29:22)
- vina_ermagan says if we assume that they don’t
have control over the SFFs, then the last SFF has to do the encap —
maybe without NSH header — to the final EPG (tbachman,
17:29:55)
- vina_ermagan says both SFC and GBP both have
the ability to do the encap (tbachman,
17:30:35)
- paulq says post service chain — the packet
comes out w/o NSH (tbachman,
17:30:36)
- normal routing/forwarding has to handle
that (tbachman,
17:30:43)
- in the case of GBP, the overlay has to handle
that (tbachman,
17:30:52)
- in the case of non-GBP, it would just be
handled by normal routing (tbachman,
17:31:01)
- mickey_spiegel asks who puts the overlay there
at the last SFF — GBP or SFC (tbachman,
17:31:16)
- paulq says that after the service chain, SFC is
done (tbachman,
17:31:23)
- alagalah says that’s his understanding as well
- and vice versa — inside the service chain GBP doesn’t care
(tbachman,
17:31:39)
- vina_ermagan says that there’s an overlap there
tho (tbachman,
17:31:47)
- paulq says not really — the SFC rules are
written on NSH; post-decap, it’s just an IP packet (tbachman,
17:32:12)
- alagalah would request SFC #1, gets back ‘you
will send packets to start chain on OVS-1, port 1, NSH header
A’ (tbachman,
17:34:11)
- alagalah says at the end, coming out of the
tunnel openflow switch 2 port 2 — also getting this from SFC
(tbachman,
17:34:31)
- alagalah says we also get the reverse
path (tbachman,
17:34:40)
- alagalah says we’d then need to key our flow,
setting up the tunnel such that a packet is going to SFC#1, we need
to get a VxLAN encaped packet with an NSH header (tbachman,
17:35:29)
- alagalah says coming out, we just have to take
the packets from the tunnel and strip the headers (tbachman,
17:36:07)
- alagalah says SFC has now knowledge that GBP is
using a VxLAN overlay (tbachman,
17:36:27)
- vina_ermagan says then GBP must have control
over wherever the last SFF is to be able to get back the
packet (tbachman,
17:37:17)
- alagalah says yes, the OVS would be
shared (tbachman,
17:37:51)
- alagalah says that’s everything that’s SFC is
matching on an NSH field (tbachman,
17:38:40)
- paulq says for the POC, let’s assume it’s NSH
end to end, which allows us to avoid the multiple writers
issue (tbachman,
17:40:03)
- vina_ermagan says that sounds good — the
handoff is local from SFC to GBP on the vSwitch (tbachman,
17:40:22)
- mickey_spiegel says if we’re doing the local
handoff on the vSwitch, then some details need to be worked
out (tbachman,
17:40:43)
- alagalah agrees (tbachman,
17:40:46)
- alagalah says we’d have to address conflict
resolution in order to do this, which would be tough to do in the
given time frame (tbachman,
17:41:05)
- ebrjohn asks what the difference between
alagalah’s google doc and the wiki page that repenno started
(tbachman,
17:41:36)
- alagalah says this is just a brainstorming
document — prefers gdoc with history, etc. (tbachman,
17:41:52)
- alagalah says the gdoc is also meant to address
phases/goals beyond the POC (tbachman,
17:42:19)
- ebrjohn says he’d like to clarify a point in
the email on the conflict resolution — we do need to make sure that
GBP and SFC are each dealing with their parts of the network
(tbachman,
17:43:20)
- paulq says yes — for POC, there’s a clear
separation of who controls which switch (tbachman,
17:43:33)
- paulq says the only constraint is that the last
SFF has to deliver an IP packet to GBP (tbachman,
17:43:49)
- ebrjohn asks if we should consider a “no encap”
option (tbachman,
17:44:08)
- paulq says it does mean that the forwarding is
“clean” on the intermediate switches (tbachman,
17:44:29)
- paulq says when you have a clear identifier in
the packet then it’s easy to avoid any conflicts in the
middle (tbachman,
17:44:55)
- ebrjohn asks if we want to address the
different phases in alagalah’s document (tbachman,
17:45:22)
- alagalah says we should push the transparent
case to a later phase (tbachman,
17:47:33)
- ebrjohn asks if there’s anything else we should
mention for phase 0 (tbachman,
17:48:19)
- ebrjohn asks folks to review repenno’s
document (tbachman,
17:49:18)
- tbachman is looking into providing an example
patch on how SFC and GBP might integrate for the POC (tbachman,
17:52:13)
- ebrjohn asks if we should continue this in the
GBP meeting (tbachman,
17:52:31)
- folks agree to continue in GBP (tbachman,
17:52:38)
Meeting ended at 17:54:37 UTC
(full logs).
Action items
- (none)
People present (lines said)
- tbachman (97)
- odl_meetbot (5)
- alagalah (2)
- abhijitkumbhare (2)
- ebrjohn (1)
Generated by MeetBot 0.1.4.