 Hello, welcome to the last session of today and I believe it's one of the last sessions of the official summit tomorrow is only design summit as I Think so I'm happy you are still here and it looks like you are not asleep yet So it I know it will be a hard thing with the last presentation here But we try to keep you awake somehow so We want to bridge a little bit between opnv and open stack because we have the feeling that people have different understanding of what is the difference or Maybe opnv is just using open stack and adding a little stuff to it So we want to explain what is the relation there and is opnv just an extension there So, my name is Ulik Lieber. I'm working with Huawei now since five years and based in Germany and I'm Huawei's TCT TSC member in In opnv I Am also leading a project called octopus for the CI program in opnv and my colleague here Yeah, I am Prakash Ramchandran. I started with only on the opnfe when it started in somewhere in September of last Year and then last last year and then what we have done is Prior to that I was working with open stack So I have some background not so much but a little and I know it's a open stack is an ocean all of us know and just to start Open platform for network function virtualization that is opnfe acronym and open stack It's a stack which is open. Hopefully for infrastructure. So given these two terms. I will just reflect we are in open stack Mitaka and If you look at What essentially we are looking is we have domain we have infrastructure domain for compute for storage for networking and then We these are all kind of a microservices that is they live by themselves They don't share anything with anybody the only way to share is Through what they retain in their databases each one of them has individual databases and They through the API calls this full API some Fun who is a tenant from? From the provider side or from a user side having an ID calls and Once they call then the question is how do you tie these together? So there is a messaging bus, which is the AMQP So that messaging bus is what helps Deliver to the dashboard user or a client user that is command line user And so you can see the compute the storage the networking object and Your block storage and object storage both are required Essentially these are services. So essentially you are looking at infrastructure as a service running from dashboard and being Having a idea and endpoint to serve through the keystone, which is the global module and images are there So basically this is what the total services the pain point here Is networking everybody knows it? So let me move forward and then bring you to Besides open stack. What is there that we are talking about and I think I should hand over at this time Having told what is open stack to earlier? Yeah, go ahead so when we use open stack now for telco environment, then we come with additional requirements and some very famous one we put here on the on the table and Many people know this f-cup stuff, which is fault management configuration management availability management or automatic scaling Things performance management security management software management all these things Are something we might add if it's not there, but partly they are there So these requirements are something From telco is coming. So is opnfv now just an extension of open stack adding these requirements? I believe not opnfv's role is a midstream project So we do different things in the top Bubble there. I put a list of different components. We are integrating in open stack in opnfv So To build a platform for nfv We need many things We need open stack and it's it's the biggest component we have there, but we have also different things like virtual switching like SDN controllers We have three different SDN controllers. We work with here We have all these Linux components. We have dptk So These are components Building blocks We do integration to build this platform So opnfv as an integrator is that all I think for the industry It's a great achievement that we in an open source activity. We do the integration of these Components and everybody who started to integrate an SDN controller with open stack knows that this is a an effort that takes some time and Providing that from an open source site from one project to the community is a big achievement So is opnfv just an integrator? So we have two Inputs there already One are requirements new features that we want to add to existing upstream projects The other thing is getting multiple of these projects together and Then on the right side I put another thing. I named that use cases and scenarios so This combination of different upstream projects we built to a platform We have different variations there depending on what we want to do in a certain network So sometimes we don't need all the components or some customers have Certain SDN controllers they like better than other SDN controllers So we have variants and these different Ways to build up the platform we call scenarios and test those So this is another input of opnfv so opnfv builds up from these different dimensions and with that provides Services to the ecosystem or it provides an input to an ecosystem of the telco services And who are those it's the telco carriers the network providers But it's also the vendors that can use all these services and depending on what we talk about Whether it's the scenarios or the integration work or the requirement work. These are different benefits People will have when they start using opnfv So what we want to do in this talk is we walk you through these different things and we show you In an high-level way, what is it opnfv has provided for that? Yes So many of you will know that we just or it's not so just anymore We have provided the Prama Putra release. So all our releases are named after rivers and in the alphabet so this Prama Putra release was provided in the beginning of March and We put a theme on that release. We call it lab ready. So opnfv is still a pretty new exercise So we don't have the feeling that the project is already mature enough that you can build a customer network on that In the telco area it will take a little bit longer until you really can do that But it's ready to be used in labs for whatever The carriers or the vendors need to do in the lab to to build their products When we look what we did We have 30 projects of opnfv that participated in that release other projects were not ready yet And it was about 165 developers doing the work. So it's not that small exercise what we did there and Based on these scenarios what I showed you We provided this release so we have four different installers That help you to bring the the software on the target machines We can deploy it in a virtual environment or on a bare metal This is all also against scenarios and we have three difference here SDN controllers To be used in this network Open-core trail is a little bit behind still, but it will be there And we have optional features that you can deploy or you can leave them out to have more stability in the network so the release Because of the the nature of opnfv is not just a software package that you can download It's different things that play together to help you in this lab work So it's of course a target platform the software image you will run to deploy your cloud and In this software image, it will be the NFE I and the VIM components according to the etsy architecture But people have told that so much in in this summit, so I won't do that again In addition to that we provide deployment tools These are these four installers that you can use to deploy easily this integrated platform and Everybody who tried from scratch to install open stack on his computer speed in the virtual or on multiple servers It's a pain to do it. It's a lot of work So we try to make that easier for our users By providing standard configurations Where you can deploy that so you can use a configuration guide to see how you have to set up your network and this this environment And then you can just download and start this thing and it will deploy on this standard Configuration will be which is a lab with five servers To payload nodes three control nodes so we can have a of open stack and a jump server Which is a server doing this deployment? Then in addition to that we provide a test framework So you can start this to find out whether the stuff you have deployed on these machines. It's really Doing something useful and you can check whether every change you do Will still work and you you can run vnfs on that and of course documentation, but that should be Obvious that we have to provide these things an Additional thing opnfv provides which we don't put into the release But it's also Something very useful for the communities. We work on requirements so Requirements are not part of a target platform, so they are not directly related there. They will be Something for the future. So we provide also requirement documents that are not part really of a release and an additional thing is We have community labs So when you want to test something in the telco area you usually need more computing power than you have at home So we provide labs from the participating companies That community members in opnfv and maybe also other people can use to do these nfv testing And I think these are different dimensions of the stuff opnfv is providing to the community so opnfv is structured into different projects and these projects are Very different in the in the nature because opnfv provides these different dimensions of things to the community So in on the left side these blue Hexagons they stand for projects that are somehow working on the integration on or deployment or Documentation so all these infrastructure part The light green ones in the middle these are testing projects So they concentrate on either creating test cases make them a run on that or doing performance testing or Defining benchmarks and things like that everything with what just connected with testing is done there and The big part on the right side then these are requirement projects Where we look at these telco requirements? F cups and so where we started and try to bring them into the upstream projects where still things are missing and In this list I showed to you that open stack is a big part of that It's ten projects that where open stack is the major target of these requirements But they are also very a Very big number of projects where open stack doesn't play such a role So some of them work on multiple ups in projects or some of them work on ups in projects besides open stack So here you see opnfv is much more than just an extension of open stack It's a it's a mid-stream exercise with which provides Multiple dimensions there. So let's go into a little bit more detail of that So the these requirements when we start with the requirements part those address different areas Open stack or other Ups in projects are working with a big part. You can call its maintenance It or on M like some telco call it. So it's all about How do we do with the with falls? How can we recover from them? How can we report them? How can we repair? How can we make the the platform more? Resilient so they they will not break when they're when the bug occurs and Think and upgrade and things like that Another area is networking So Prakash all already said that networking plays a major part and we have more a little bit more on that Another thing is orchestration and the interfaces We need to provide to the higher level components in such an architecture and The third area where we have to work is that a telco network is not placed in a single data center and During these summit we had many talks About how we can work with multiple sites or multiple instances of open stack Working together or breaking a data center into micro data centers. Just the session we had here before So these are the major areas we are working to Maybe I should explain this RIS there It's our help so resilience and then you have the Availability and service ability so this other area to just to bring it on Also to this picture. So I'll put it this obviously we work with carriers and Verizon has brought this in service assurance is mostly Intel is trying to address and Resilience and availability availability is always 999 so and resilience is being able to get You fall down you get back. So that is a resiliency. So whatever happens failures. You have to come back That's the resiliency and the fcaps as already only mentioned This is important. That's why it say again and again this fault Configuration then accounting performance and security, but then people have different different a they will add availability and then as they will add software so essentially covers the Carrier greatness which is required Based on the various functions. So these two are there and we have been debating internally within the OPNFE as well as within the Organization we vendor community and what should be our target? We started At Arno, which was more focus on continuous integration because we knew a lot of software or upstream We just have to pull them put that together and yet build a Reference which will be interface compliant with it see NFV and so on so forth, but software So development is independent of that they run faster than the standards can bring in so we have to Ensure that we are able to cement that and so we struggled in the first Arno release then we came to Brahmaputra Brahmaputra. Now we have 21 scenarios as he mentioned yes for for each of the three Installers plus the eight or nine which is based on fuel and fuel is part of obviously open stack But it doesn't have all that we need so we are enhancing it then in Colorado. We decided. Okay, what do we do next? So we have a theme of net readiness we would prefer to have a carrier grade but Even network is the biggest pain point and the challenge is how to make it network ready and you had sessions from AT&T on glue on and proton and we are still dreaming but we need to Engage with open stack to ensure that we plug in at the right place and then we have it's cloud which is another Evolving piece in the three tier cloud instead of three tier services now We are looking at those things and finally we are also looking at mobile because Mobile is king and video is king. That's what the traffic says are so our roadmap options are there But this is not final. This is what we are thinking and we want the feedback from all of you to help us understand how we go from here So now we are at the halfway stage almost we will just drill a little bit detail and I will ask Uli to start with the maintenance So just as an appetizer here for the these maintenance topics I think the most successful projects we have broad and the area of Requirements is the doctor project and we had several Talks during the summit so I don't need to talk a lot of it But this diagram is from doctor and shows how this notification flow goes through such in platform until the application gets notified about a Fault that has happened on a platform The resilience resilience and upgrade are the other areas there. Let me say a little bit about the upgrade area What we need in the telco area is that an upgrade will be seamless for the services So we don't need an upgrade that we will keep everything alive at any time That's not what we need We might even be able to shut down open stack for five minutes or ten minutes as Long as all VNFs can operate during that time an upgrade can be seamless for the services of Course after a certain time when we shut down the controllers for an hour that will be a problem and If we shut down the SDN controllers it the problem will occur earlier than these five minutes so we have to look into the details where do we need which Downtime requirements during an upgrade and How can we make sure that we can deploy new services very quickly in such a network? So there the future will provide many many new Requirements and mechanisms to be able to do so Networking so networking and networking is computing We have this debate going on for years and years and decades and decades. So at this time When we start in the NFV reference architecture, we started in the WIM Which is open stack? 80% let's say started with that, but then networking is the key. So we had to introduce SDN controller as part of WIM in fact the first starting itself we started with ODL and then We said that at the same time open stack was evolving You remember they had ML 2 ML 3 plug-in ML 2 plug-in ML 2 then plug-in then driver then agent and so you got empty number of pieces to tie together and our drivers and the plugins plugins Also were thrown out. Sometimes they were brought in in the neutron. So we had a Advanced services and so we bring it from Havana onwards. We have seen a lot of things happening and in this moving working environment, we had to plug in the Onos and ODL so we started with ODL and we brought in immediately onos because we need to have duality and multiplicity of so that We are able to work with the plug-in and the use cases actually drive them So one simple use case if I say SFC service function changing that is one of the key pieces of At least for carrier grade and for the telco So we talk of middle boxes like we talk of firewall We talk of load balances and we talk of gateways for mobility gateway for fixed network then VEPC's there are number of VNFs or virtual network functions composed of multiple VMs or single VM and OpenStack in its own Neutron doesn't give all the flexibility we need and this SDN controller will be able to do the underlay Stitching for those service function changing based on flow classification based on the open flow Whatever is supported and default is OVS, but default OVS doesn't Help us deterministic way of doing which is the requirement for telco grade and therefore we had to also there is a sink as in Data sink issues and so these are all getting resolved one by one using these use cases So we have SFC VPN L3 VPN L2 L2 VPN we were able to accomplish in Brahmaputra But in Colorado we are shooting for L3 VPN and BGP VPN and Then IPv6 so some of the networking features which will help us service function changing from Odl side of it as well as from the ono side of it. So these are the main features we want to incorporate and these are Essentially we are trying to include them We have scoped them lot of gaps are there and it's not the question of gaps because OpenStack comes from totally IT centric data center centric is based not so traffic whereas for us it is more of a control plane separation data plane separation and data flow that is the throughput for the and latency so there are characteristics which are different for use cases and therefore Network is the biggest pain and hopefully we will be able to address with the moving target still We are seeing lot of disruptions here like today. We heard about we use support example Port support we use for SFC for service function changing for network SFC whereas now we are talking of Router instead of top of rack switch. We are talking of top of rack routers as Sean Collins put it use heard him And that's another disruption we are seeing so It's a work in progress. Hopefully we will be able to manage Expectation and deliver what we need based on the requirement and we will keep engaging with OpenStack in this So next one. Yeah, so Now I will talk of only we want control plane data plane separation. So data plane. We want to have fast or throughput of the flows of different applications being either tunneled or straightaway Pass through now that there is a limitation and because we want to build it on the cards That means the server we don't want to change But essentially we feel that there is a need for hardware abstraction To be able to separate out and that hardware expression. This is D-PAC is data plane acceleration Proposed by China mobile and they are working on it they have They want to do the packet processing cryptographic processing and encapsulation decapsulation for the Traffic which is in the data plane and therefore you find that there are multiple ways of doing it and that is their way of One way of doing either through hardware or without a bit abstracted with the drivers and then on the other side we have Fast data stack which is Cisco's way of if you look at the bottom it says Linux FDIO and DPDK. So DPDK is your compute acceleration and on top of it you have the Vector processing packet processing so that you can buffer and over the DPDK and then Provide some kind of a API call honeycomb API what they call in the ODEL and that being used by the ODEL based implementation for fast Data parts so this is what is happening So there are multiple ways of acceleration and we do provide diversity and there is no one vendor that will dominate We have all vendors coming together and trying to standardize it. So this is a good piece of Global collaboration I put it because we have China mobile which is almost eight times that of AT&T's so you can see the need for them and There are a lot more to it. Okay. I have not even characterized power consumption data center like cooling. So there are many aspects to it so next one will go and I'll just briefly say we have not reached the performance. These are only requirements But we know that given the best practices what we can do on a bare VM We should be able to do it in a VM as well as in the cloud and the goal is to make sure that You cannot do everything on a Lab ready or a field ready. Just you have to have best practices Policies and all those will come eventually to get that there So I'll just skip this and now This is important piece this piece you can understand that when you have a network and You have underlay and overlay of course so underlay you can have some kind of a graph and you have ionic which can do the stitching of that at the lower layer by installing the Computers and storages and all that in traditional way But when you go to overlay then you have to look of a service being translated into graphs Generally, most of these service can be modeled as a graph and so once you do model Then how do you drive the policies through them? How do you create them? How do you get so which one comes first is computes comes first or network comes? So somebody says no not bound API of compute is important. Somebody says no not bound of SDN is important That's not true. It's a combination there off. So everything can be under control if we map our use cases as a models and as a logical model of the nodes and the links and Then drop from top the services based on the flow classification and run through the graphs Help them drive that through that and the underlay There is also tunneling aspect of it, which we have to address right now. Just not traffic It's not packets is the bunch of packets put together with a signature of something We call it as s-flow flow and so those are with IDs and all so that's why models and policies are important because you have local and global Separation and you can have policy at the global level and you have to fan it down to local based on the What you call the infrastructure domains or domains as we call them or maybe segments or there are many many definitions There I'll leave it to that slices and all so we have to implement those policies And this is what we have projects in opnfv like models. We have vnf forwarding graph. We have sfc We have Copper we have plenty, okay So a lot of these are brought into picture to be able to provide that kind of service. I Think I'll give it to yep orchestration is is another aspect very important Orchestration is something which was added to the scope of opnfv only last November So we are still on the beginning on adding this because in the beginning when opnfv Was created is it was limited to the nfvi and Wim layer Or bnf v orchestration we have to start at the OSS bss layer and come down then to the data centers and infrastructure The important thing there is to the life cycle management For both the vnfs and for the services we run So I won't go into many details But we will do similar things with the orchestration like we do in other areas going for the diversity So the last bullet here says mono options. So everybody knows there are different exercises now addressing Orchestration in open source we can start with what we do in open stack with a attacker project But there are also other exercises like the open o We had a presentation this morning and also the osm Project which runs under etsy to create in open source Solutions for orchestration Also in this area in it's important to have all these automation tools We're also open stack is contributing a lot too. So this is an area opnfv still has to address And we are in the beginning to define the projects which will work on these topics So So on this slide I summarize once more after we talked so much about the requirements We are working on the the other aspects. We are providing to the communities pre-integrated scenarios and In the Colorado release we have defined 21 scenarios all of them need to be tested in our release flow in our CI pipeline To be able to to define these and provide this to the ecosystem and just two examples What could a scenario be? these acronyms there OS no SDN and so it means we combine open stack in This scenario no SDN controller no additional features, but H a so we do an installation of open stack in in High availability mode or as another example Combine open stack with honors as SDN controller Install service function chaining features and the H a so all these combinations lead to different things there then the testing framework Our testing framework is based on the functions provided by open stack there But we enhance those 10 percent rally and robot Then the community labs also some information there a few companies and there are more who provide these Labs on their premises and have the community work on this or under CI pipeline on that Another part that the opnfv plug fests and the first plug fest will happen in two weeks Not so far from here a few hours. You can go by car in Colorado There we will start with interoperability testing of the opnv members bringing products bringing solutions of opnv and test portability there It's a new exercise and I'm really looking forward to to see what we can achieve with these plug fests and Then we are working on a compliance and certification program open stack has That as well, and I think for opnv. It's even more important to work on this compliance and certification We have another slide on that. I think it's the next one. No one more Yeah, let's do the compliance and certification first We call this project working on that dovetail, which means something which Binds thing together like this It's still a working progress So what we do there is we create a framework for this certification and verification of components That work with opnv or in an NFV environment to help the multivendor aspect of NFV to become reality Yes, yes, this is similar to what we have in Opus stack here rev stack and and we run the capability metrics by modules by We have additional modules like hypervisor and we drive the use cases We also have levels like we'll have core mandatory optional, etc. And then be able to work on testing by module and ultimately certification can be by Yes, you can ex be exercised by the opnv project. So Or we can exercise it by 30 party or by the vendors themselves So this is still under discussion how we can do this certification program So I think that's it. Let's skip the back to the CIC shortly. Yeah so Opnv provides a CI pipeline for our development So we do automation with Jenkins Like it is done in the CI programs in open stack it on one side. It's easier because We have one built for for for the opnv target platform Whereas in in open stack all the modules builds that themselves then they are combined and on the other side We have the complexity with these scenarios and have to run that so we run a master Jenkins program on the Linux foundation lab who is our host and We run then the slaves on all the community labs and have the dashboard as you know it by that The CI pipeline will go then to to continuous deployment via the installers and automatically Install the target platform on one of our labs and also Automatically start the testing so it's it's a pipeline from the the CI the city and the continuous testing And also the this tooling will will be a good thing to provide for the community So save you for the time. We have a few minutes. Maybe you can ask a couple of questions I think at the end of the day everybody wants to run out we are Ready to shoot to the door, but you are welcome to ask a couple questions and we will answer And yes, thank you for the interesting presentation My question would be our opnfv versions and open stack versions related in any way They are so Prama Putra is based on the Liberty release Honor was based on Juno. So we take take stable releases from the upstream projects For this integration effort because it it provides for us a more mature basis to do this integration work The Colorado we have not done the final decision But on my opinion it should be based on meter car And there is a six months cadence anyway, so we follow the six months cadence to accommodate As many upstream as they can be accommodated at this table release Thank you. Go ahead If you have any local changes Will you release it? Will you upstream it before you release it or will release have some local opnfv specific change You know some more neutron or some other project. We always work first upstream upstream first is a principle there Quick question about orchestration Does that I saw an FVM there, but does that end scope of the orchestration piece? Orchestration what is it question? So you said orchestration was added recently to orchestration Yeah, we we do support Several tools like you have puppet ship ansible salt which are just the tools in themselves, but then actual orchestration occurs Based on what the OSS BS is right Yeah, my question is about VN FM. So are you gonna have yeah VN FM options? We have right now. We are not finalized Packer is one of them and then the possibilities are open Open O is a project which has been brought under Linux Foundation opnf. He is also a project under Linux Foundation What it means is that open O which is initiated by China mobile and AT&T They are working to get some kind of a orchestration at the level of VNF manager as well as NFVO Orchestrator in reference model the tacker deals currently with VNF manager and that deals with the TOSCA and templates and all and there is a requirement which is fanning up saying that we also need to have Yang modeling and TOSCA modeling combination may be top-down bottom-up types So SDN controllers currently are supporting only the Yang modeling like ODL Onos is still trying to get native Yang modeling. So you are going to have some Uncertainty at this stage as what eventually the question was was about VMware there. So we are open source Because you talked about orchestration being added recently. I just wanted to clarify that the VN FM function Yes, so the VN FM function is in scope there Yes, VN FM. Mano has the scope for VN FM NFVO both the service side resource side and at the Vim layer we deal with all the SDNs Can we expect anything in Colorado? Colorado we are not seeing anything at this time except for testing using tacker testing using Domino there is a policy template distribution And we we don't have certainty at this June time frame come open all gets ready By the time we go into D at that time. We'll have a decision. There are no decision What is the process for engaging the open NFV like Three group is it a mailing list and yeah, I see how do I ask for features or report bugs So we have mailing lists. We have a wiki where you can can read and find out what is working There in the different projects we have IRC meeting. It's pretty similar to open stack when the working mode there Do you ever have working conference sessions like? Yes, I mean it depends on the project some some projects don't have meetings at all Some projects have pure IRC meetings some projects have voice meetings some have the meetings to time zones because we are all over the world Okay, are those listed on the Yes, you find it in the wiki wiki.opinofi.org And then and and you don't need to be so so opnve has a bunch of member companies But you can join the work in opnve even if you are not a member Okay, so it's very what is the difference between someone who is a member individual member versus platinum silver and associate member So platinum is the top tier and then you have middle tier which is the silver And then we have the associate members like universities and all and individual contributors. They are most welcome any time Yeah, so you if you have anything call us or send us a mail to Info it we are happy for everybody So so everybody wants to to contribute there of course needs to sign some contributed agreement And if you are in a member company, it's easier because then your company already has signed for you And you just has have to click something simple in the internet at linux foundation sites but if you are Doing that on your free will or so without a company behind that you just have to read a little bit more Okay, thank you very much. I think we have Come to the end of the day. You don't have any questions or I think we should wind up gracefully and Hopefully go and enjoy your show once more the email addresses. So if you want to yeah, if you want anything Here is the website wiki.opnve.org as well as you got dub dub dub which is called the Marketing material wiki is the right place for you to Developers and then you have our email IDs So feel free to contact us if you have more questions or want to join Thank you and enjoy your evening