15:03:29 <serverascode> #startmeeting operators_telco_nfv
15:03:30 <openstack> Meeting started Wed Nov  2 15:03:29 2016 UTC and is due to finish in 60 minutes.  The chair is serverascode. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:03:31 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:03:34 <openstack> The meeting name has been set to 'operators_telco_nfv'
15:03:41 <serverascode> #topic roll call
15:03:45 <serverascode> hi all!
15:03:59 * PerfectChaos waves
15:04:07 <ad_rien_> o/
15:04:10 <ad_rien_> Hi
15:04:23 <GeraldK> Hi
15:04:35 <jamemcc> Hello
15:04:54 <serverascode> ok great, we have the most ppl ever :)
15:05:07 <serverascode> #link https://etherpad.openstack.org/p/BCN-ops-telcom-nfv-team
15:05:21 <serverascode> ^ that's the session from last week at at the summit
15:05:34 <serverascode> #link https://etherpad.openstack.org/p/ops-telco-nfv-meeting-agenda
15:05:39 <serverascode> ^ that's our agenda page
15:05:57 <serverascode> I only have one real thing on the agenda right now, feel free to add to it if you have other things to discuss
15:06:04 <serverascode> the one thing was our list of potential projects
15:06:19 <serverascode> anyone have any comments/suggestions/ideas before I go into that topic?
15:06:57 <jamemcc> No - good organization Curtis - lets push forward
15:07:44 <serverascode> ok
15:07:49 <serverascode> #topic Potential Projects
15:08:27 <serverascode> so the idea was to follow the large deployment teams lead and fiind a medium to long term project to shepherd through the openstack ecosystem
15:08:43 <serverascode> they worked on getting network segments into neutron and were pretty successful
15:09:04 <ALUVial> o/  (back to lurk mode)
15:09:29 <serverascode> basically they found a common feature or issue and worked on it over time
15:09:40 <serverascode> so we have a list of things from the summit at that link above
15:09:52 <serverascode> anyone see anything in particular that jumps out as needing to be solved?
15:10:26 <GeraldK> NFVI benchmarking ref.platform got +4
15:10:36 <GeraldK> Rolling/live upgrade has +3
15:10:49 <serverascode> right there was a lot about benchmarking in that list
15:11:14 <serverascode> how about the people in the meeting today? what are your issues or requirements?
15:11:15 <ad_rien_> Is there a specific use-case that can serve as a guide line ?
15:12:01 <GeraldK> upgrading is something we (DOCOMO) would be interested in
15:12:34 <ad_rien_> We can identify a lot of challenges but with a common theme it might be easier to focus and make priorities between those challenges
15:12:38 <GeraldK> currently, we skip releases and would like to improve on this
15:13:03 <serverascode> right, and it's not really possible anymore, afaik, to skip releases
15:13:39 <serverascode> ad_rien: what's a common theme? like "benchmarking" is a theme? is that what you mean?
15:13:55 <ad_rien_> From all the challenges I can see on the etherpad
15:14:02 <GeraldK> #link http://specs.openstack.org/openstack/openstack-user-stories/user-stories/proposed/rolling-upgrades.html PWG user story on rolling upgrades
15:14:28 <ad_rien_> I get the feeling that some are rather general (i.e. upgrade is an interesting challenge but not specifc to the NFV WG If I'm right)
15:14:34 <PerfectChaos> Rolling upgrades strike me as something that are generally going to be of interest to anybody
15:14:43 <PerfectChaos> ...yeah, exactly what ad_rien_ just said =3
15:14:52 <ad_rien_> So this is my point.
15:15:31 <ad_rien_> There are several challenges, all look really interesting but it might make sense to identify the ones that are really specific to the NFV use-cases
15:15:40 <GeraldK> in contrast to other openstack users, Telcos might prefer not to have very frequent rolling upgrades.
15:15:48 <PerfectChaos> I guess the thing to consider is - if it's a general problem there are more likely to be other people elsewhere working on them.
15:16:12 <GeraldK> got your point.
15:16:13 <PerfectChaos> Whereas NFV-specific problems are less likely to have people who aren't use out there driving to get them solved
15:16:23 <ad_rien_> PerfectChaos:  yes exactly my point
15:17:02 <serverascode> ok, it seems like there is a bit of a consensus that working on upgrades might not be NFV specific enough
15:17:37 <serverascode> that leaves us with benchmarking as a theme, at least in terms of what has been mentioned today
15:17:46 <ad_rien_> It might become critical/specific but later.
15:18:13 <jamemcc> Just to give it a little more insight I think the possible more NFV related use case comes through with the massive cloud assumption
15:18:15 <PerfectChaos> Well, if the main problems NFV/Telecoms folks are all general problems, then we shouldn't just discard the idea of looking into those. It's down to what those people will find the most useful, right?
15:18:53 <jamemcc> If you go with the assumption that the NFV/Telcom brings with it the massive cloud - then the upgrade of massive cloud has addtional challenges
15:19:29 <jamemcc> That said - I think somethign closer to Telcom/NFV is more appropriate
15:19:30 * PerfectChaos nods
15:19:38 <ad_rien_> Maybe we should start by simple actions/challenges and then increase the complexity step by step.
15:20:15 <ad_rien_> ForI agree with jamemcc remark but this mean that first you should be able to have an Openstack deployment accros distinct locations
15:20:35 <ad_rien_> s/ForI/I sorry for the typo
15:21:01 <ad_rien_> Is there a specific NFV use-case that can help us in our exchange ?
15:21:29 <jamemcc> @ ad_rien yeah - agreed if I follow - So back to defining the reference implementation?
15:21:36 <ad_rien_> yes
15:22:23 <serverascode> by reference implementation...what does that mean?
15:22:33 <serverascode> is that kind of like what OPNFV is doing?
15:22:42 <ad_rien_> What is the OpenStack deployment architecture if you want to address NFV use-cases? How can you instantiate such an architecture?
15:22:48 <serverascode> ah, ok
15:23:25 <serverascode> At work we have been trying to define NFVi, specifcally around features
15:23:46 <ad_rien_> is it deployed in one location or across several locations? where are the compute nodes that host the VMs? The control plane? …
15:23:54 <ad_rien_> serverascode: NFVi ?
15:24:23 <ad_rien_> i for Infrastructure ?
15:24:30 <serverascode> my impression is that under the ETSI definition, OpenStack is NFV infrastrcture (NFVi) and a VIM
15:24:47 <serverascode> VIM = virtualized infrastructure manager
15:25:01 <ad_rien_> yepp
15:25:02 <ad_rien_> thanks
15:25:20 <ad_rien_> so that also something that can help the OpenStack community
15:25:36 <serverascode> and it's hard b/c it's really a spectrum of requirements
15:25:41 <ad_rien_> I get the feelings that our words (i.e. from the network community vs the distributed computing community) differ sometimes
15:25:42 <GeraldK> ETSI NFV003 defines NFVI: totality of all hardware and software components that  build up the environment in which VNFs are deployed
15:25:56 <GeraldK> so it's more than Openstack and also includes the hardware
15:26:16 <serverascode> ok thanks GeraldK
15:26:23 <GeraldK> it also could have multiple VIMs
15:26:53 <ad_rien_> but you can have different ways to deliver such a NFVi
15:26:59 <ad_rien_> see in the pad line 37
15:27:24 <ad_rien_> either a brokering approach where an orchestrator operates several VIMs or a bottom up approach
15:28:04 <ad_rien_> where OpenStack becomes cooperative enough to be able to natively federate distinct data centers/locations
15:28:21 <serverascode> interesting, so are we discussing reference implementations right now? kind of?
15:28:32 <PerfectChaos> I think we are, kind of.
15:28:38 <ad_rien_> I would say yes (at least from my side ;))
15:29:01 <GeraldK> I believe Telco's want to stay in control so prefer the brokering approach
15:29:09 <PerfectChaos> The first step would presumably be deciding how deep the reference implementation goes
15:29:12 <ad_rien_> There should be different scenarios, a first action can consist of identifying them
15:29:49 <serverascode> I almost feel like in order to get any work done we first have to define at least one or two reference implmentations
15:29:56 <ad_rien_> GeraldK: we work with Orange Labs and BT and the choice is unclear. We are trying to see what are the pros/cons of the different deployment strategies.
15:30:00 <serverascode> so that we at least know what we are talking about to one another
15:30:14 <GeraldK> and then, how different would this be to the reference platform developed by OPNFV? Or, how could we collaborate with OPNFV in it?
15:30:58 <serverascode> I was hoping, at some level, that this group could work as a bridge between OPNFV and OpenStack Operators doing NFV
15:31:10 <GeraldK> to work on the main topic of "NFVI benchmarking" we may not need our own reference implementation, do we?
15:31:25 <ad_rien_> serverascode:  yes this makes sense
15:32:00 <serverascode> no I don't think we need to make our own definition, but we'd have to have some common language to do work in this group
15:33:13 <serverascode> ok, so we are about 30 minutes into our hour long meeting
15:33:48 * ad_rien_ is looking on google for information on the reference architecture of a OPNFV deployment
15:35:06 <serverascode> I think we have discussed three themes so far: 1) upgrades 2) benchmarking and 3) reference implementation definition
15:36:01 <GeraldK> ad_rien_, maybe a good document to check current status of OPNFV is http://artifacts.opnfv.org/opnfvdocs/colorado/docs/documentation/index.html
15:36:04 <ad_rien_> Maybe a stupid question/comment but is there a link somewhere to clearly understand the differences between OPNFV and OpenStack in terms of internal mechanisms
15:36:36 <GeraldK> not sure I get your question. OPNFV is using OpenStack.
15:36:43 <ad_rien_> ok
15:37:00 <ad_rien_> Let me reword, what does OPNFV bring?
15:37:20 <serverascode> that is a question I hear quite often :)
15:37:23 <ad_rien_> from my understanding
15:38:06 <PerfectChaos> Basically, they bring together various pieces that combined make an NFVI. One of those pieces is OpenStack.
15:38:30 <ad_rien_> OPNFV enables administrators to program the network active equipments (in somehow, an advanced Neutron component)
15:39:11 <ad_rien_> so the reference architecture can be several openstack instances and one OPNFV to deliver a NFVi ?
15:39:20 <GeraldK> there are several goals: integrate various upstream projects, act as intermediate body between ETSI NFV work and OpenStack, join forced to push new features to upstream projects, ...
15:40:13 <GeraldK> ad_rien_, OPNFV builds a reference platform including OpenStack, ODL, and others
15:40:45 <ad_rien_> ok
15:41:02 <GeraldK> there was a nice handout on OPNFV in the Summit. let me see if I can find it online
15:41:15 <ad_rien_> but If I'm correct the working group focused on NFV within the OpenStack ecosystem?
15:41:32 <ad_rien_> so do we care about ODL and others :-) ?
15:42:18 <serverascode> I think we do have to be aware of SDN systems like ODL, and OVN
15:42:44 <serverascode> certainly some NFV deployments will rely on functionality that only SDN can provide
15:43:03 <serverascode> but not all deployments will need that. Then there are projects like neutron-sfc
15:43:18 <serverascode> oh and vlan aware virtual machines which was just blogged on superuser
15:43:36 <ad_rien_> serverascode:  sorry for my remarks, it was a bit provocative. I agree we should be aware but we should also focus on OpenStack, shouldn't we?
15:43:42 <serverascode> #link http://superuser.openstack.org/articles/four-opens/
15:43:54 <serverascode> ad_rien: no worries, not provocative at all
15:44:36 <jamemcc> Agred focus inwards toward OpenStack and represent combined knowledge interest of OPNFV and SDN
15:44:50 <jamemcc> and ETSI
15:44:55 <ad_rien_> AFAIK, there is not reference architecture for delivering a NFVi with OpenStack, maybe we can try to identify how we can progress on that point?
15:45:02 <PerfectChaos> So... since there are a -lot- of different variables which could vary wildly between use-cases even within NFV/telecom use, we'd have a hard time picking one reference implementation which would be consistently useful, I think.
15:45:05 <ad_rien_> at least as a starting step?
15:45:50 <GeraldK> #link http://go.linuxfoundation.org/l/6342/2016-10-11/3jhnwr/6342/158966/OPNFV_WhitePaper_Paving_Way_OpenSource_NFV_101016.pdf WhitePaper on OPNFV
15:45:58 <PerfectChaos> Since people were interested in benchmarking NFVI, would it instead make sense to be thinking about frameworks/infrastructure with which we might be able to test/verify/benchmark different NFVIs?
15:46:08 <GeraldK> I couldn't find the handout, but the figure are also available in the whitepaper
15:46:12 <serverascode> I do think service function chaining (SFC) will probably be a common feature, and with neutron-sfc you don't technically need a 3rd party SDN controller
15:47:02 <serverascode> so we could do a reference arch that supports neutron-sfc
15:47:37 <ad_rien_> We (at Inria) are interested by performing such evaluations
15:47:42 <serverascode> also there is some interesting work around "neutron router flavors"
15:47:44 <serverascode> #link https://specs.openstack.org/openstack/neutron-specs/specs/newton/multi-l3-backends.html
15:47:50 <jamemcc> I think to define reference architecture and combinging wiht what is said about focusing on OpenStack - w are saying what openstack components are included and in what arrangement to get basic NFVi working.
15:48:11 <ad_rien_> We have access to the Grid'5000 testbed so we can perform experiments in a quite easy manner
15:48:25 <ad_rien_> jamemcc:  +1
15:48:43 <jamemcc> And from my perspective this would be across at least 3 separate clouds though perhaps form testing and benchmarking we ought to pick a higher #
15:49:13 <GeraldK> neutron-sfc is also being discussed in OPNFV (https://jira.opnfv.org/browse/VFNGRAPH-3) but I am not aware of any details on this project
15:49:55 <GeraldK> IMHO this topic requires some discussion on collaboration with OPNFV. I am sure they would be interested on some collaboration.
15:50:12 <serverascode> ok, interesting, so I'm feeling a bit of consensus around some kind of "vanilla" or "minimal" NFVi reference architecture, perhaps including neutron-sfc?
15:50:21 <ad_rien_> +1
15:50:47 <ad_rien_> I think I should be able (or at least give a try) to add Orange folks in the loop
15:51:25 <serverascode> ok, so we have about 8 min left
15:51:46 <serverascode> where should we go from here? do we agree this is one area to explore?
15:51:54 <GeraldK> the goal should be to get some benchmarking or similar, but not to have "just" a reference architecture.
15:52:27 <jamemcc> Agreed
15:52:34 <jamemcc> and likes it
15:52:53 <serverascode> ok, and do you think it would be benchmarking some kind of "vanilla" or "minimal" NFVi based on OpenStack projects?
15:53:08 <serverascode> so neutron-sfc for example
15:53:09 <ad_rien_> First step can be a reference architecture, Second step a scenario that inclue neutron-sfc  (maybe we can push such a scenario in Rally) and Three perform some evaluations?
15:53:21 <PerfectChaos> See above about benchmarking frameworks - what about, say, coming up with some standard set of benchmarking tests for those evaluations?
15:53:53 <PerfectChaos> (which could subsequently be run against a different NFVI to see how it performs in comparison, if one was so inclined)
15:53:54 <GeraldK> what I mean is: create a set of (standardized) benchmarks and test those on a reference implementation. but the focus would be the benchmarks.
15:54:24 <PerfectChaos> Right. Sounds like we're on the same page, then.
15:54:31 <GeraldK> agreed
15:54:42 <serverascode> cool, there is also a performance working group that we could team up with
15:54:45 <ad_rien_> PerfectChaos:  can you point once again the link regarding the benchmark ?
15:55:09 <serverascode> that would fit in well with the user committee wanting more communication between openstack working gorups
15:55:19 <serverascode> we could collaborate with OPNFV and the performance team
15:55:45 <serverascode> ok so can I say something like we agree that:
15:55:46 <ad_rien_> We are involved in the performance WG at Inria. If I'm correct here we do not really talk about performance but more on functional benchmarks ?
15:55:52 <PerfectChaos> ad_rien_: Hmm? I didn't have any links on benchmarks?
15:56:05 <serverascode> ah ok functional vs performance
15:56:25 <ad_rien_> "PerfectChaos: See above about benchmarking frameworks " - see above ?
15:56:41 <PerfectChaos> My assumption was that it would include both functional and performance benchmarks...
15:56:45 <ad_rien_> ok
15:57:40 <PerfectChaos> ad_rien_: Oh, I was just referring back to my previous comment "would it instead make sense to be thinking about frameworks/infrastructure with which we might be able to test/verify/benchmark different NFVIs?"
15:58:06 <PerfectChaos> To which I think the answer is "we should be thinking about both".
15:58:13 <serverascode> ok, can I say something like: "We agree that this working group will develop and perform benchmarks on a NFVi reference architecture which we will determine in future meetings"
15:58:36 <PerfectChaos> That seems like a good point to start from
15:58:48 <GeraldK> what would be the timeframe for this? and would we focus on this one only or allow a second topic?
15:58:58 <GeraldK> but definitely a good starting point
15:59:07 <serverascode> I think we'd have to come up with a timeframe in future meetings
15:59:14 <serverascode> we could certainly work on more than one thing
15:59:21 <serverascode> only about a minute left
15:59:26 <GeraldK> sure. but roughly? 1 year?
15:59:50 <serverascode> probably something like that, at least a couple releases
15:59:55 <serverascode> sorry I have to end the meeting!
15:59:56 <GeraldK> okay
16:00:01 <GeraldK> bye
16:00:10 <serverascode> #endmeeting