 All right, welcome. Hi, i'm randy levinsaw with cable labs and i'll be moderating this panel today We're going to be talking about nfv management and orchestration Using open stack i'm really looking at this a lot from the telecommunications and service provider perspective We have two great panelists with us today i'll start with Uri is there If you can introduce yourself sure. Hi Good afternoon everybody While we are only two we'll try to give you as many different perspectives and opinions on The subject at hand today Personally i'm with intel data center group As a cto of the data center network solution group, which is a group inside the data center My Charter covers everything from man or through open stack open daylight v switches other data plane technologies to to the server active in multiple open source communities like open daylight Open o that Was active till the end of last month also working in the itf on service function training co editor of nsh and multiple other things but We'll talk about it. I guess in the next 40 minutes Yeah, and lingley It's working. Okay. Uh, hello everyone. My name is lingley. I work for china mobile and I have been Involved extensively in several of the open source initiatives around nfv domain including opnv Uh, open previously open now and now on app and and also internally and Also leading the efforts for stn and nfv trial network in china mobile in our vision Nova network will be our next generation Network and practice and we're also working really actively on the open source orchestration side Excellent. Thank you. I said i'm randy with cable lab. So we do research on behalf of the cable industry specializing kind of the software side of stn and nfv Very active with opnfv and Open stack for quite a while So we kind of yeah excited Just as a quick high level detail slide. I'm sure everyone's seen this a thousand times, right? Is everyone seeing kind of the etsy mana model raise your hands. Yes. Yes Good a few people didn't raise their hands, but right. You know, we really have kind of the nfv orchestration layer Up here, you know the vnf manager and the vim and the vim and nfvi is really open stack And we're open sex playing so we're going to be talking about, you know a lot more about How do we manage this infrastructure? How do we scale across multiple data centers? Just for a little context and then yuri has a couple slides to kind of kick things off to Give a little more context here and then We'll go into a few questions and we do want this to be a very interactive panel So start thinking of your questions now and we'll get those going Yeah, so so maybe just to get you guys a little bit Hopefully excited One of the phenomena that we have in the industry is a little bit of a tower of bubble I'm using this for quite some time One of the hopes I believe many people attending this event and open stack has Is that open stack could be some sort of Grand unifying Piece of the puzzle that we all want to see used in the industry from telco perspective Many of the initiatives and we we all obviously are going to mainly refer to Open source as it's easier to refer to that to those ones as there are no limitations Are really focusing on using few elements like Open stack for the cloud management since telco is a little bit different Then cloud there are a few things that come to play so That that just one one more second So the tower of bubble is really how do we get some convergence and specifically also not only the software pieces that we're using But also data models What we're doing on this slide is really Trying to put side by side the way innovation is happening in cloud versus the way It happens in telco and how open stack can play an interesting critical role in Accelerating the pace of innovation in in telco. So if we are looking at the traditional Telco comms model It used to be and it still is active and if you missed an opportunity to travel to niss in france, it's a great location Don't don't miss the next one They come over there. They meet for some time create some standards at see in this particular example and And those standards Take two to three years They get codified in some sort of information model And then the vendors take off from that point They create their own proprietary product and and the dashed line that you see Is where those products over some time after user experience and after requests from operators Takes that long and and it potentially affects and triggers and and creates some innovation in the next cycle Cloud is a very different model In cloud you actually start with some software because I have some idea some innovation And sometimes there's going to be some agreement To keep the slide simple. I use this they call this data model, but really it's most likely an api And in the rare case there's going to be also some sort of standardization If we look at containers right now people are Hopefully aware of stuff like oci cncf So when the industry feels the need they go and and and have some agreements around api And that would feed back sometimes into this this model obviously it's much much faster It it works very well as we all know it creates software that scales and How do we really get? To use the benefits of this model in a telco environment and this is where What we are doing in the last few years is really trying to fuse together the two things if We look at what we need to do in telco and the end of the day We all remember that it's in many countries regulated industry. So it needs to be anchored into standards But we could still start with iterating on software So we could have open source Communities and lingley and myself have been working in the last I don't know 18 months or so in different communities specifically to Mano and to telco and mano is in most cases targeting Open stack and then we could play with data models in order to achieve broad industry agreement In parallel to working with standard organizations and even standard organizations now Hit the call and they understand they need to move faster, so you had Phenomenas that the industry have not seen like all the standard organizations really coming together in one place In order to figure out how they work among themselves and how they work with open source communities So that is really kind of the logic behind why the industry is working In open source why that works for everybody and the important role that open stack plays in this Now in the next few slides one example that Led to work we did in in open o so the key assertion on on this slide is to say In order for an ecosystem as complex and as I think broadly Used as as mano where the telco industry renovates itself and and goes into Uncharted waters of everything virtualized And and cloud models What was happening in the industry? Maybe up to 18 or 12 months ago or even closer to present Was that people were thinking about oh, I need to create a mano and mano to me represents It's a telco push button where I get automation From whatever network service I need to deploy to the server that actually supports that bear in mind that open stack Does not have the notion of what the network service is does not have All sorts of business regulation life cycle requirements that the telco industry wants to see So while we box them together in mano That that is only one piece of the puzzle the key idea here is that for this to take off Technically and from business and from changing the way organizations work We need to think about the problem holistically We need to make sure that they are going to be workloads that are Adjusted to the manos keeping in mind that there are many manos That would mean that as a vnf vendor I have to do lots of work I have to go and do what's called onboarding For different environments that is taking way too many cycles from these vendors right now And essentially slows down the whole industry. So one example of work We recognizing the pain of Vnf onboarding that is talked around in the whole industry What we did in uh, if yeah, we're not going to spend time unless people are going to come back with questions What we did in open or and I had the opportunity to serve as the vice chair architecture over there If you want to move to the next one and and just just So the idea here was to Help the industry move forward By a concept that has much broader implications than the way it is used here for telco Where the application or the workload Has an active meaningful interaction With the orchestration as well as with the infrastructure So the idea here and we are we in open. Oh We have this code ready. This is actually one of the projects that Was accepted into own up for those who don't know open. Oh Was a very successful community initiative started by china mobile that Linkly is Is a part of but was an open source community with many participants and That community falls into own up that is even larger Community open source you could look it up. What? This is about So the vnf can say I need the following resources That trickles down through manor Trickles down through open stack to the infrastructure that enables the infrastructure The ability to give you some boundaries some guardrails around what you need to do What resources you Want to have so we could protect sla. We could improve efficiencies and at the same time Help eliminate the noisy neighbor problem again open stack would look at the world as I simply need to go and and place some Workload here and there would not have those higher Level concepts would not know how I'd like to scale my workload when demands is changing What is even the concept of network service? So you have two elements here What are the resources that I want and what is the feedback that I'd like to get so I know my workload is healthy And with the combination of these two elements I could see first of all reduction in the Onboarding time and energy that people need to do I could actually if I have these Type of extensions to the data model and I have industry agreement to that I could see that my workload is going to work on multiple environments and so on so maybe stop over here All right. Yeah, let's get lingley involved a little bit now You know we've talked about challenges at scale china mobile is you know the largest Telecom operator dollars just number users you talk through some of the challenges that you're having Today at with you know scaling your services and what you hope to achieve with NFV Yes, actually internally we have two separate teams working Like NFI or open stack one is dedicated to the scenarios of public and private cloud Which you will hear on I think frequently from them the scale, you know Like problems and issues and practices that they're they're doing when when they're doing our commercial deployment for those cloud Applications, but we have also a separate team working on NFV scenarios. Well, you would be in vision Not so much of the scalability issues of the platform itself because we think that you know in our net we have this like vision that our Core networking structure will be replaced by a two layer of of new type of data centers Which is managed by I think cloud management systems like open stack, but All those clouds or telecom clouds we call it like telecom integrated clouds Small sized, you know regarding to typical like deployments for public clouds For example, we might have only several hundreds of nodes in each of the sites in and we don't you know Encarned so much like scalability issues, but we wanted to enable as I mentioned earlier in another I think panel for 5g like applications and arrows So in those telecom scenarios, I think high reliability and ultra low latency And also multi-talency and things like that would be the key requirements for cloud infrastructure And if folks have questions just come up to the microphones on either side and we will address them So one of the you know other topics, right? You know onap is now the new project coming forward and we've talked about you know interoperability a lot and How's the vnf verification process going and the maybe talk a little bit about the vnfsdk project under onap Yes, as as Uri has done I think a quite great job in the introduction And actually we are kind of like borrowing that practice of vnfsdk into onap And currently I think it has separated in your like work items Last week when we are doing this kickoff for project proposal for release one And on the least I think we will have a like automatic test Framework and for all those onboarding testing and inside the service service provider domain for the service design before the service design before it can Be aborted to the catalog and also aside from that previous to that we have we are planning to providing also SDK tools for the vnf vendors who can do their self service like Testing certification before it can be you know the vnf can be packaged to a third-party marketplace Or to be you know taken into the service providers labs for their service design Slide on this if you scroll down one. Yeah, I think we have one from the open now Also in addition to that we were also working on a set of like guidelines for onap community and I think Important item from that group is vnf guidelines and also vnf management guidelines Which we'll be talking about specifications of the data models that were just you know Mentioned and stressed out for example using heat Tosca for the data modeling of different vnfs And how we're going to do vf packages with the software image and all the artifacts to Actually onboard the vnf and do the testing Excellent So if if you guys look at the slide like this this this represents the the project that We were doing in open. Oh, you could identify kind of With four vertical boxes, but the three left ones are Wrapped around with this yellow line So if you take from the left to the right on the left side, you see the tools that lingley just mentioned and this is about And in onap there is still work going on on the data model agreement, but Potentially it could be or in open or it was Around the idea that we have open source That you could use in order to create the toska blueprint as an example that was demonstrated publicly The key concept here there may be two key concepts of interest one is simply to reach industry agreement Is valuable for everybody? Let's pick one and run with it The the other one is is that It is not affecting the application itself We are really talking about metadata meaning and and The the case in point is that there was a public demo we did in mobile world congress vnf vendors came with vnfs VMs for us in open in open stack In two hours, they were able to create This metadata given these tools I would not say it was 100% functional or anything But just to give you a feel that this is relatively simple And it is not about changing the way you develop your application It is about pretty much annotating it in a way that allows the orchestration To have a clue what you are trying to achieve And the middle section is about if we achieve this kind of industry agreement with This type of data models. We could really create some sort of an api data model agreement that ushers in For the telco industry a cicd model that we are talking a lot about in open stack in in containers environment Where you have a situation where a developer on one and one side of the planet is fixing a bag or is adding a feature An operator on the other side could download it immediately It creates an opportunity for financials and business models around that with With a marketplace, which is We are working With china mobile in in onap that is an opportunity that is being discussed over there as well And what you see on the right side is simply that it's modular. It could plug to a given manual orchestration and Actually the key leading manual orchestrations right now are all using open stack, which is good news for for our community And nfv comes up quite a lot I think this speaks to one of your slides about complexity of deployments There's kind of two camps. There's those that talk about nfv a lot and they bring up a lot of use cases around that and the other camp believes that kubernetes networking is so simplified and that's one of its strengths Can you speak to any other use cases or how you may think that kubernetes would be a better system to run some of these things? Yeah, I think maybe you want So I'm going to represent first the right Hemisphere of the brain then the left one and and so you will have multiple opinions here. It's going to be rich enough There are multiple questions that are included in in the way you presented that so We could look at that Maybe starting from the pace of evolution and innovation in in the telco community in general And I think it is fair to say that there are multiple POC's proof of concepts that are popping up here and there in the telco industry, but predominantly today it is still Around virtual machines and it is the public goals that multiple operators have are Again predominantly around deployment that is using virtual machines now I believe that what we are going to see as an industry is some sort of A coexistence and in some cases where One one good telco example for that could be something like vcp that that stands for virtual customer premise equipment, which in a little bit more plain english means Instead of Telco operator having to Send a technician with a truck with a piece of equipment that is going to be placed in your location And every time we are going to change protocol or change an application or change a use model We are going to have to do this again and install it and connect it and tune it and all of this We are changing this to Some sort of a white box that has many cases open stack in it Virtual machines and now the issue of upgrade or issue of replace Becomes turn on this virtual machine turn off that virtual machine So you're using open stack in a in a cloud way the reason that example Is connected somehow to containers and lightweight is that many Times operators would like to have close to the edge of the network meaning close to us as users They would like to have some mini data center point of Presence many names for this me see whatever you want to call this And I want to control multiple Enterprise environments from that location that I have the control plane for that meaning having something that has your personality policy I have for you All sorts of other things that are specific to you or you could think of a mobile user In their car Driving by 5g multiple models like that. That is much faster come and go situation This element of the application could easily be implemented in containers so You know multiple other elements that telco would like to deploy Would be deeper in the network. They could be Using existing equipment of virtual machines. So here's an example of coexistence and Working on containers a little bit the question of container networking even container integration with open stack many many moving Targets many moving balls right now. So I think it's way too early to call a specific direction We are looking at the way to have orchestration deal with them such that they are going to be side by side Actually, if I may add I I would like to see orchestration to be agnostic To whatever the underlying technology would be used in for different use cases If they are to be coexisted side by side within a common infrastructure We would like it to be you know to be more shared You know in the in a way, which we do not see physical right now, so probably Adding another like type of virtual containers. It seems to be Adding complexity to the management and also orchestration We wanted to you know to have the especially I think operational or business like Service logic to be agnostic to that. So I think we were I think we will have discussions within open No, and also there is already I think our requirements projects at opnv to to have that discussion of what are the requirements for Container environment what it means for cloud management and also the orchestration management and We are following that up and see if we can you know We're right now doing some of the pocs, but in a separated manner We haven't seen anything that can coexist in a hybrid manner and And solve the resource like issue very efficiently And I am personally have these doubts how these two kind of going to co-exist and to be managed in a very high efficient manner because to me that that seems to be like Separated resource pools managed by different managers infrastructure managers And there's no way currently to talking to a common like orchestration management layer and all these are you know like Blocking issues for months to think about a hybrid or side-by-side experiment Yeah, we're seeing quite a bit that as well in the cable industry looking at using containers and kubernetes closer to the central off or closer to a centralized data center And even doing that for some of the ease of configuration on top of the ms But as you get closer to the customer site The more expensive it is to locate and manage that equipment So we really need even if it's just a five percent improvement in performance That's actually very important as well as less moving parts, which means less less truck rolls Which means kind of better service for everyone So are there other questions of the audience? Please come up and ask Um, so, you know, we've talked about on app a little bit I know for the first release, you know, we had the meeting last week with the developer meeting and we have um Three use cases that we're targeting A link would you like to talk to those use cases that we're using to drive the first release of on app? Or at least are proposed? Yes, very briefly. Actually, there were Proposals on the table from the tsc discussion and they are yet to be approved by the board And the proposals were the virtual firewall the simple one which is already running, I think in over open ecom cope base and we wanted to enhance that adding a more like Close-loop automation flavor to that and and and also serving as a like reference implementation that everybody can get access and just to um Deploy in your environment very quickly and test all the infrastructure and requirements to be set for on app deployment And the other two are more complicated including one for residential virtual cpe And the other for volt including uh, virtual ims and virtual e pc for those two complicated use cases new use cases we're looking to um include Recruiting both open source vnf's in as as well as commercial vnf's so that we can really to you know to address those of the doubts from from the manual group Saying that if or not, you know, like open source or just straight a project can resolve a really Like realistic telecom use cases And just just to kind of also address your your question from before And and connect it with The kind of a little bit of an intro slides that we threw in here So when you start looking at this kind of more complete use cases like voice over lte that lingley is mentioning You become very sensitive to the sla you become very sensitive to The performance as an example so you you really get yourself concerned with What is the throughput I would get from from a server? What is the right? Network architecture that I'd like to have on the server and as well as in the data center in order to be able to scale Rapidly everybody is driving home around five o'clock if you're lucky, maybe you're working longer hours. I don't know but then Then that creates a little spike as an example or if you have a sports event or whatever So how do we create those? Such that when I put that component that is doing the ims that is really real time oriented Nobody would like to hear stuttering on on Is or her voice connection? How do we guarantee this? How do we make sure that those vnfs have been correctly placed? They have no Interruption from other vnfs on the same server from From the network every chosen the right architecture as an example in open stack today you could We we have contributed to the community something that we call an ends platform awareness that allows Open stack implementation and the man knows example zero are using this To be able to say hey for this I want s r i o v and going to kubernetes and containers There are multiple folks who say yeah, I'll use that but since I'm trying to have a use case like that I better have s r i o v or I would have other forms of offload In order to get to that performance that I need the beauty of doing that with open stack Is that the code is there in order to automate that so I could really From the application point of view. I could simply say This application would like to use dpdk or s r i o v or this flavor of a v switch And code is being injected automatically to the compute node in order to connect that to neutron to nova In order to make sure that the image is properly connected to whatever v switch you are using To do this say with a linux bridge if that is one of the interesting ways The container community wants to use that Requires some work And you get yourself into the conversation of Features security Performance how many cores I'm using for communication and so on and so forth All right, great. So we only have like two minutes left. So I won't ask another question I think this was a great session and really appreciate everyone Yes talking especially about what we're doing with on app in the future And some of the opnfv work. I don't know if you have a closing statement But I would move to you know as we're maturing from you know ideas to actual real life running code Well, I I think that There is Maybe maybe a good Way for those of you are interested into What's ahead and what's in the future is We have a session on on thursday about talking about northbound and southbound the apis For the the whole telco stack the conversation and you could find it As in the context of an hirvana stack, but honestly, this is a little bit broader conversation that talks about Detention and opportunity between getting industry agreements on api on data model Where does the beam start where does the beam and where does man or start? What kind of agreement I need on on a data plane? And I believe that that is very appropriate for the open stack community to have a strong voice in this That would be important. Well, thank you very much for your insight. You're yin glee