#opnfv-sfc: opnfv-sfc-weekly
Meeting started by tbachman at 13:59:32 UTC
(full logs).
Meeting summary
- agenda (tbachman, 13:59:45)
- http://ircbot.wl.linuxfoundation.org/meetings/opnfv-sfc/2015/opnfv-sfc.2015-07-15-14.02.html
Minutes from last week’s meeting (tbachman,
13:59:50)
- https://app.gotomeeting.com/?meetingId=611478989
(tbachman,
14:00:29)
- https://global.gotomeeting.com/join/611478989
Goto Meeting link (ebrjohn_,
14:02:12)
- http://lists.opnfv.org/pipermail/opnfv-tech-discuss/2015-July/003920.html
Email outlining today’s agenda (tbachman,
14:02:36)
- action items (tbachman, 14:03:27)
- ebrjohn_ has created 5 sequence diagrams
(tbachman,
14:03:50)
- https://docs.google.com/presentation/d/1gbhAnrTYbLCrNMhMXin0lxjyg-7IHNPjrlBTIjwAzys/edit#slide=id.g5b68d7718_60_54
document for sequence diagrams (tbachman,
14:04:01)
- paulq says he’ll have the arch document
sometime tomorrow (tbachman,
14:04:16)
- lmcdasm says tacker could be used as a VNF
manager, but it’s a bit immature — might be better focused on
SFC-specific pieces (tbachman,
14:05:04)
- ebrjohn_ would like to get closure on the
mechanisms for SF/VM creation, and how to get started with the VNF
Manager (tbachman,
14:07:30)
- ebrjohn_ says it would be good to get a few
people working on the OPNFV setup scripts, too (tbachman,
14:07:58)
- http://lists.opnfv.org/pipermail/opnfv-tech-discuss/2015-July/003916.html
Email showing OPNFV-SFC has a git repo (tbachman,
14:08:44)
- lmcdasm says it would be good if we can be very
specific with the building blocks and dump the pieces into the
repo (tbachman,
14:09:15)
- SF and VM Creation (tbachman, 14:10:08)
- ebrjohn_ says that there are 3 different
options for SF and VM creation — feels that option #2 is the way to
go (tbachman,
14:10:48)
- ebrjohn_ asks for opinions on the options for
SF and VM creation (tbachman,
14:12:18)
- ebrjohn_ says the difference between the 3
options is the mechanism for VM creation — ODL, operator calling
into VNF Manager, etc. (tbachman,
14:13:49)
- paulq says he’s not opposed to option #2, but
feels it’s not sufficient — there’s another case where the SFC
infrastructure would need more of a given service, do we expect ODL
to signal back to the operator? (tbachman,
14:14:31)
- lmcdasm says there are two different cases:
operator creation on-demand, and another with automatic scaling of
the infrastructure using some metric. In the latter case, the
signalling is back to request creation of more VMs (tbachman,
14:15:12)
- Manuel R (MR_Sandvine,
14:15:30)
- alagalah says there’s another consideration
which isn’t just the horizontal scaling — you have a bunch of SFFs
using a scheduling algorithm. There are a few cases where you have
bottom-up and top-down. (tbachman,
14:16:13)
- lmcdasm says the addition of having the SFF
creates another element that has to be scheduled around.
(tbachman,
14:16:31)
- paulq says the point is that there are multiple
requesters or consumers of the VNF Manager/operator (tbachman,
14:17:07)
- lmcdasm says we could build this out starting
with scripts that will run each of the use/test cases (tbachman,
14:17:39)
- lmcdasm says he doesn’t want to get stuck on
the word “operator” — maybe “Vector” or “Trigger" (tbachman,
14:17:59)
- lmcdasm says tacker doesn’t have much of a
defined interface as of now, so we may want to start with the
scripting (tbachman,
14:18:40)
- someone recommends using the MANO terminology
for these things. (tbachman,
14:19:29)
- ebrjohn_ asks if folks agree with starting with
option #2 (tbachman,
14:20:21)
- lmcdasm asks if between operator and VNF
manager covers all the inputs we’d want (tbachman,
14:20:38)
- ebrjohn_ says the point of the slide is not
meant to cover the entire SFC configuration (tbachman,
14:20:55)
- ebrjohn_ says once we’ve decided on this, we
can go work out the next set of sequence diagrams (tbachman,
14:21:41)
- ebrjohn_ added a slide for automatic VM
creation to cover that use case (tbachman,
14:22:20)
- AGREED: Group will
start with option #2 for SF/VM creation (tbachman,
14:24:25)
- bouthors asks if we should take care of whether
a SF can be shared, possibly between tenants? (tbachman,
14:25:18)
- regarding SF creation scaling, we need to
address this, but with a lower priority (ebrjohn_,
14:25:39)
- paulq says that shouldn’t matter at this level
— as long as SFC can associate it with multi-use, that should be
okay (tbachman,
14:25:43)
- bouthors was interested in the case where a
dedicated SF can be shared (tbachman,
14:25:58)
- lmcdasm asks if it would effect the geometry of
any of the VMs (tbachman,
14:26:23)
- Automatic SF/VM creation is something we want
in first release, but not until we get option 2 completed
(ebrjohn_,
14:26:26)
- paulq says not necessarily — the premise
(generally) of SFC is you present one interface into the topology,
and use the metadata to distinguish (tbachman,
14:26:49)
- How to get started with VNF Manager (tbachman, 14:28:56)
- someone mentions their company has a VNF
manager that they’re looking to contribute to the OPNFV
project (tbachman,
14:30:18)
- lmcdasm says the operator trigger will start
with one case — run a script that will talk to a server side
listener (python listener called VNF manager), which will call the
Nova APIs to create a VM from a template (tbachman,
14:32:48)
- afredette asks if there’s a scenario where
there’s a request to create a chain comes from OpenStack
(tbachman,
14:33:47)
- lmcdasm says OpenStack wouldn’t take actions on
their own — they’d do so based on external input (tbachman,
14:34:07)
- afredette says the use case is that a user
decides traffic from a given tenant should go through a given
service chain (tbachman,
14:34:25)
- lmcdasm says the trigger could be from anything
— the VNF Manager could be anything that implements the OpenStack
APIs (e.g. even a simple server-side script) (tbachman,
14:35:11)
- ebrjohn_ says the short answer is
“maybe/possibly”, b/c there’s a lot more involved (tbachman,
14:35:50)
- alagalah says the RSP is based on the EPs
involved (tbachman,
14:38:52)
- alagalah says the EPs are the consumers of the
chain, and not elements of the chain itself (tbachman,
14:39:18)
- bryan_att says he doesn’t think there’s a
project in openstack specifically to create service chains or
VNFs (tbachman,
14:39:51)
- someone asks to explain what NFVI is in the
picture (tbachman,
14:40:39)
- lmcdasm says its the NFV Infrastructure
(tbachman,
14:40:55)
- Stuat Mackie says in the original architecture,
the VNF Manager includes connectivity between components
(tbachman,
14:43:21)
- lmcdasm says for the context of our demo, ODL
will set up the chain (tbachman,
14:43:33)
- Bryan Sullivan (bryan_att,
14:43:34)
- OPNFV-SFC Setup Scripts (tbachman, 14:45:33)
- ebrjohn_ asks what’s necessary to get SFC up
and running in OPNFV (tbachman,
14:45:43)
- lmcdasms says lets agree on a platform
first (tbachman,
14:45:52)
- lmcdasm asks if we’re going to use ARNO, and
which deployment method (e.g. Foreman) (tbachman,
14:46:10)
- lmcdasm would like to create a quick system
list, along with the target, and what goals we want to accomplish
(e.g. create chain XYZ) (tbachman,
14:46:37)
- alagalah says that ChrisPriceAB sent out a
suggested architecture on this (tbachman,
14:46:50)
- ChrisPriceAB says we can’t use ARNO vanilla,
but use ARNO with Lithium, and will have to use the GBP piece in
order to get SFC up and running (tbachman,
14:47:12)
- afredette asks why we need GBP with SFC
(tbachman,
14:47:23)
- bouthors says we need GBP to say who’s
consuming the chain (tbachman,
14:47:36)
- ebrjohn_ says we don’t have an openflow
classifier in SFC yet — just one in iptables (tbachman,
14:47:49)
- bouthors says even if we have a classifier, we
need something to say that we want to consume the chain (tbachman,
14:48:04)
- trozet asks if that’s just an openflow
rule (tbachman,
14:48:20)
- ChrisPriceAB says using the GBP, you define the
service chain, path, and RSP, which pushes the rules to the
switches, which aren’t correlated with openstack unless there’s
something in the middle to create that correlation (tbachman,
14:48:51)
- trozet says you can use a VNF manager (e.g.
tacker), then push a simple flow to push traffic to that
chain (tbachman,
14:49:07)
- lmcdasm says there has to be a second call to
SFC to make that happen — sounds like there are two modules we want
to try: iptables, and GBP (tbachman,
14:49:38)
- alagalah asks how to handle the mapping
multitenancy, address namespacing, etc. (tbachman,
14:50:05)
- trozet says this is just an initial step
(tbachman,
14:50:06)
- alagalah points out the GBP integration already
works (tbachman,
14:50:15)
- ChrisPriceAB says it’s possible to do it in the
VNFM — he’s not sure he wants too much up above the VIM (tbachman,
14:50:38)
- paulq says it’s faster to get GBP up and
running (tbachman,
14:50:54)
- bryan_att is looking for help in getting the
GBP + SFC up and running (tbachman,
14:51:06)
- alagalah says they demo’d the SFC integration
at the IETF hackfest (tbachman,
14:52:13)
- ACTION: ebrjohn_
provide an initial list of software needed for OPNFV SFC
(ebrjohn_,
14:52:32)
- ChrisPriceAB says he also got it running at the
hackfest — does require some digging, but works (tbachman,
14:52:46)
- lmcdasm says there are a lot of blades/HW
available to explore both, if interested (tbachman,
14:53:20)
- ACTION: ChrisPriceAB
to set up a meeting at OPNFV hackfest during ODL summit to look into
using the demo (tbachman,
14:55:04)
- ebrjohn_ would like to try to find a room at
the summit to have f2f OPNFV meeting (tbachman,
14:55:50)
Meeting ended at 14:56:23 UTC
(full logs).
Action items
- ebrjohn_ provide an initial list of software needed for OPNFV SFC
- ChrisPriceAB to set up a meeting at OPNFV hackfest during ODL summit to look into using the demo
Action items, by person
- ChrisPriceAB
- ChrisPriceAB to set up a meeting at OPNFV hackfest during ODL summit to look into using the demo
- ebrjohn_
- ebrjohn_ provide an initial list of software needed for OPNFV SFC
People present (lines said)
- tbachman (101)
- alagalah (9)
- ebrjohn_ (8)
- ChrisPriceAB (6)
- trozet (6)
- collabot (4)
- bryan_att (2)
- MR_Sandvine (1)
Generated by MeetBot 0.1.4.