 Good afternoon. Thank you for coming. My name is Hain Wang, I'm Danny's school so come from Huawei and then a friend and the customer Alex Suset from Waterfall. So today we come here to talk about The topic that we are very interested. We will try to encourage you guys to interest So convergence of IT and the CT infrastructure So we know that We're all here for OpenStack. So OpenStack Has winning the support across industrial as a future IT Framework or control panel. What are you called? And the NFV as initiated by telcos Is widely endorsed by telco industries for their future blueprint of the infrastructure. So this is lots of It's too trend, but there's lots of similarity between them and it will be good for us to combine these two together So this is what this talk about So my talk will be breaks three parts. We're talking about the relevant industrial trends So why this is? Transformation is going to happen. What's a technology driving force and then we're talking now We talked about from telco operator perspective. What really they want From an M3 from OpenStack from the converging from structures and and finally and most importantly we'll talk about our experience in terms of challenges and opportunities for this convergence, so This talk intended to get OpenStack community interested so we can achieve the OpenStack mission to make OpenStack as Your big your biggest operate open source Cloud platform to be everywhere. So Okay, so in this trail and this is all the old news hardware trend is software defined and so a Software defined really has a virtualization for underlying components to making the accessible Through API. So what really means they make hardware? API driven elastic infrastructures What they do they abstract the hardware and pulling them together and automated management configurations for API's so specifically to Technologies virtualization and the scale out. So this has been down very well on server That's when we are pioneered this and now it's doing a network and storage So what they do you want to you're able to programmatic provisioning Since the separation of decoupling hardware from software So with the separation you were able to place a move the workload everywhere and You provide more operation efficiencies through automations. So that's all we know what a virtualization bring to us and it's gonna happen Network and the storage and that they are flattening the structure So the more pool and a fabric and a commoditization using the standard components. So that's the trend so driving force is really We let a software decide What they want? that has been That were very different of 10 20 years ago because the Scarcity of hardware Everything is decided by hardware you optimization around hardware and right now hardware has doing them well We're so good that they become less important in terms of optimization because it every Morals law make them every 18 months double the capacities, but software doesn't so now the lazy one become dominated so software You've all more rapidly than hardware and infrastructure should be flexible to meet the needs So that's a hardware trend an open stack. So cloud we know early practitioner like Amazon has approved the value and we know there's a Seq infinity challenge a change the IT we know the challenge of presumptions that we have regarding computing not necessarily Expensive is good and the commodity hardware can do much better job. So the open-stack mission I just repeat that we produce a ubiquitous open source Cloud computing platform that will meet the needs of a public and private cloud and then they are architecturally they're proven by the public cloud offers like Amazon a rack space and then made it accessible to the masses For use for our community commodity hardware another interesting trend in open-stack and in general software that innovation is happening under the open source and we see this So big data for cloud and many other technologies. So open source is no longer Follow it's become the leader the driver And We also know that the cloud adoption given the value given the Need should be much faster, but they they are not they are very slow I think though for part of reason maybe technology the other reason that the options limited so we have Amazon as that definitely They pioneered a club is all credit for them But they are the only one wonders not many people in this room can participate this other than as a channel and a way I'm with the dominant enterprise virtualization, but it's a wonder driven centralized architecture So looking at a history anything for mass adoption You usually have to a disruptive way to do it and open source In this case Linux is a way to do it and we see this happen in the history We see the Linux a Unix Linux. We see the iPhone and Android so open source definitely will drive the adoptions Much quickly so NF we as part I know I just repeat what the Mission is and driving force There they try they want to replace purpose build Hardware with the standard hardware for better price performance ratio and There's a lot of reasons and I think they basically want to challenge the assumptions that Amazon challenged it before that Community a standard component can support the network of functions that telco needs It's just like Amazon saying Expensive high performance high reliable not necessarily the better IT infrastructure So the driving force for this is low cost of operations and we see telco have lots of pressure The workload has significantly changed social web OTT Bring lots of traffic, but where a few Where little revenues our friends and all technical operators know that so hardware cost Also the react react to the market and how to fast the provision in your new service and then the old network application the cycle is well long and managing and Maintain the distributed equipment is challenger without a new Platform and of course fast time to market is another business driver. So this diagram. I don't repeat this is this on a white paper and Fwe they're talking about a moving this a silo Boxed approach into the common hardware infrastructure and application on top We see that there's a lot of similarity. What are we talking about a cloud? So this is the NFC reference architecture again, this come from the white paper so you can see bottom left box the NFC infrastructure and you have a Compute network storage and then you have virtualization on top of that and then you build their apps on top of that So this is very similar. We can see the similarity between this and Cloud and so is nature. We think cloud is perfect platform for NFC and this is just an example saying one of the Telco apps on the left you really box approach you have a special build hardware box ATCA Equipment that you have the apps line up and then they have dedicated Switch fabric behind and how controller you have apps running on top that and it give you the very good performance And when you move a crowd is more like you the physical there they all share same compute the resource there and then you move those apps into the VMs and Then and then you're built on top of that So idea that application don't change you just move them there and that's the idea approach What they change that the decoupling between software and hardware? Because the VM hide the complexity of what actual Hardware is this virtualization bring to you stand as interface between any of the elements and infrastructure We know the cloud is has become sort of a new hardware provide API for your access and Telco network and a deployment is for automation rollout efficiency. Those are the cloud bring you a resource pool that can cover the net all the network elements so We're drawing another one. So if I put them there a Open stack Form that you have bottom you have open stack a cloud you can choose different hyperwise that you like and then on top you have traditional Middleware support a telco specific apps and you have a middleware that support a tradition at the IT apps OSS Yeah, and in the future. I hope this could be a new path that have abstraction to support Telco apps So this is basically the trend that we see we see the match But does that work? What's the challenge ahead? And is this what telco want? What they they are understanding So here I bring my friend Alex here and he's come from Waterfall. He's a chief cloud architect so Alex all right, so We wanted to give a sort of telco operator perspective and I wanted to share Kind of the conversations. We're having Within the business. It's not only a whatever conversation. It's an FV conversation And we wanted to put that in the context of what opens that can do Towards an FV because we we think it's very good match and we'd like to give some indications of what the challenges are and What really the perspective and the opportunities are for open stack? So I always like to start with why you know, why are we doing NFV? Hi spoke a little bit about it So basically we have three main drivers standardization is very important and It's an optimization of how we use our resources. So To give you an example today When we go in the core we have a very tight link between software and hardware. So it makes upgrades very difficult It forces us to build love resilience forces us to sometimes even change the up the hardware when we need to an upgrade and That's actually quite difficult to manage. So With this standard and with the abstraction from software and hardware we can enable a much more Clear standard and communication towards our vendor as well We can then increase resource utilization because we can start sharing the hardware between different vendors There's a lot of rules that we need to apply in order to do that in order to guarantee quality of service And that's the last piece here But we're getting to that on what open stack can do for that One of the aspect that we're very interested in is the programmatic aspect. So What are the APIs that we want to expose out of these platforms? You know, how can we Give our vendors a Clear line of of separation and say these are the APIs we have that's how you can use the infrastructure And that's the standard we like you to use So we always think, you know, develop once deploy many times is really the way forward A telco application is extremely similar between Vodafone and Telefonica and a lot of the other telcos and the quality of service as We're moving into the IT world more and more. We are starting to use it standards IT hardware We may even use open stack. You know, we always keep in mind that it is not perfect It does fade, you know service fade how this phase So it's really about having the visibility of the environment and I would get into that a bit later But in the end, why are we really doing this? Well, it's because the user experience on the telecoms today Is expanding all the time. I mean, if you think five years ago, you know before the iPhones before the smartphone movement before the applications We had a very clear demarcation on what? The core services were supposed to do but today it's evolving at a pace that we never saw before So we need to adapt new ways of being flexible of being innovative So the way we want to achieve that is well first of all, you know Having high resilience and then being a lot more distributed So sharing resources programmatic exposure once again, I said a lot, but standardized and proven technologies And then the second piece is where we really want to be is we really want to accelerate the pace of innovation We have a lot of induced resources that we could actually expose for POCs or for what I call less traditional use case We can actually introduce Very small part of the core for a very special application which in the past was extremely difficult to do And and we don't know what this more application is going to be So we need to put in place the technologies the tools the standards in order to allow For this more innovative approach for for this more open and fast-pacing So back to open stack This is a view that that we discussed and then we'd like to Communicate what you know, we feel that you can do as the open stack community. So I think there are three important focus points. The first one is performance So what we mean by performance is that? We know open stack is highly performance. It's highly scalable But we need to have more inside view of What the hardware is doing? And it's very important to come to have a sort of agreed quality of service so we cannot just expose core resources and then start sharing everything and then Having a service decrease because another tenant is using too much resources So that that's why we say it's good to build in SLA information so that we can be more flexible We can guarantee we can give extra resources to workload But when the time is here for a customer to maybe make phone call that resource has to be there for the phone call and guarantee So I think the footprint will be important for for open stacks. So How can you expand into sort of NFV type organization? Maybe expanding to discussion with GSMA. So a lot of the standard bodies around telecom. It will be very interesting to to get more involved and the resilience So how can you put in place Mechanism that are going to guarantee that if we say this application needs, you know, that amount of resources And if something is fading, how do you support that application in being a resident and maybe restarting somewhere else in the environment and really helping the application Having this dynamic relationship to the infrastructure and The last point I really wanted to make today Which is one of the discussions that I always have with the open stack people is We need to be really strict about the standard on the APIs as soon as we start to introduce APIs in these environments It's very important that We can keep using these APIs going forward that the API doesn't change so backward compatibility is very important We can expand the APIs we can evolve them The backward compatibility is prime because we don't want to have applications breaking when we upgrade Especially in these type of environments. So that's it from my side We get back to you Thanks, Alex. So Are we Now with the partner we have been experiment NFV open stack and here we share some Observations we have an experience and the challenges and problems that we think we're facing. So there's from from the Platform level we do see right out of bed. There's an obvious difference from the focus So what NFV want and what cloud current cloud provide? so computation versus connectivity so We know the cloud study is a virtualization on server side So it's nature that they're more computer centric the kind of platform Why the NFV? The network is more network centric. So this is the focus is different And then secondly that why we we have big apps in the In cloud word, but most often The workload is a small app So the cloud enable the good thing the cloud is that cloud enable multi talents with many relative small apps To share the resource the computer storage network in a more efficient way Why the network the NFV needs a scalable network function? They can serve meanings and tens of meanings of subscriber For one or two apps. So they are sort of a big apps. I mean constructed Cloud is more small apps, although it's not extreme. It's in general and then the what you like virtue and a physical word of separation So cloud really encourage a totally decoupling. So the physical word What you would don't know anything about a physical word, but I'm the network functions Under they are there's a handoff between the physical word and What you would so how you can be connected to view among them? other things that Tell is more interesting from network operations and from the manageability So the challenge we see the future will notice that as we along work on our paths first is a Cross data center hierarchy resource scheduling. So as I mentioned a telco app instance You respect a multiple data centers. It's big. It's multiple countries and one apps an operator And then not only that operator want to be look at as a one hierarchical or one Center they can deploy them and you distribute up among the different data center. So these things that are Not very obvious in the current cloud affinity scheduling and telco apps really have multiple De-coupled apps related together. So there's a special tension between their relationship and the heavy intern VM traffic They're also the high availability and policy governor redundancy Those are the very different telco because reliability Scabily they always focus on telco Then the cross version upgrade the app the upgrade On the cloud would Right now we more focus on the platform or we am never but telco is more on the they have a lot of data Planned traffic these southern some of hardware accelerators. So this make platform upgrade not idea for their case redundancies and Also, they are up focused instead of platform. So these things are the things that we know is Challenged and so the here we use a diagram to show this so you can see the problem When you move with we showed the diagram before so when you move one Boxed software model into the car model and the box model you in the back and they have a full fabric switch that guarantee the network and so the low latency of zero loss and They have controller and have dedicated the bandwidth But on the car world the best effort you cannot guarantee this a box based application rely on network of fabric That is not provided We hope the SDN can solve this problem down the road Let's think probably have best hope in telco apps and enterprise app And then I mentioned that Like a connected view of a physical and virtual. This is harder for the troubleshooting for the operations so so we Parted a few apps and work with our partners So some issues we notice on the kind of platform From the network function is a parity issues manageability performance reliability Mentability those are different and Existing operation tools are no longer meet the needs. So when you go to the virtual world the physical modeling tool will not work, but we don't have one that connect both physical and virtual world troubleshooting across physical and virtual world those are things that are missing or not exist This is all the next one problem is more as I related with affinity There were neck like the upper level support we will support a way I'm well, although we have some discussion in Open-stack community about a container Basically, you want to group resource together the support of one apps Instead of just for individual VM. This is for resource provisioning for scaling for SLA for all this stuff We need a granularity beyond the VM So you have an infrastructure have a resource container that I can group the resource when you scale you scale the whole resource as a pool instead of individuals so So we this is some sort of way we are working on that so for solving the hierarchical data center resource scheduling You can do the tree hierarchy Among data centers, and there's a route and everybody report that introduce the hierarchy on the data centers and Then the kind of know as sale probably not work. We're also for multiple data centers So we have some suggestions to extend the need to configure firewall into sales So those are the things that we think We're working on that. We want a community to also pay attention to that Because to expand the platform to the cloud or to the telcos This this challenges Important for them adopt the overcome the problem is important from the top to the car platform and Affinity schedule here. We just said We were a practice that you have in short term You can have a use a available zone and put things together and long term You probably have a better scheduling algorithm that I can imply can I can bring the policy into the scheduling? So when people deploy apps, they can just specify what they want and then engine can execute but in the time right now we use the available mechanism to do it and It's hot. It's doable. It's just harder So here I'll draw in different the diagram if you have a container just mentioned you have a container and this See the few guys doing this. This will help Adoptions that you can they can move the whole apps because idea that most up. They don't want to rewrite They want just move to the virtual world. So how you can help them behave as the city was in the box approach So if the container can similarly that environment That would be the good and then we see the discussion in the open stack Docker is probably one of the options there for the container approach and then Overall, we want is the application-centric architecture really in the technical case is more specific and On the network side is true on application level. We want to have abstraction that support the concept in NFV and the networking that you can App can actually define the network infrastructure one that can define as area on top of that and There's some work there Linux container and AWS have something like ops works Open I have a heat. So those we see them move into that direction and we just want to bring more use case For the community to work on that and Apple level HAA and data protection That's not discussed in the open stack yet. And we are more active recently pushed for HAA for app and then Apple level monitoring SRA is crucial for Operators So here we will bring more Question and challenges for you and we work on with our customer partner. We also encourage them open-stack community to work with us and to put more effort on there and then they are Obviously some signs they're looking on that direction and here. We'll just give you more evidence why this is important Tell code word and bring the big apps to the platform So this basically The slides I have and then we have all the Authors here, and so if you guys have a questions You can ask an hour. You can ask after the sessions Again, thank you Thanks, I've got a question for both of you about performance So it's interested that you were talking about the scaling across multiple data centers question Which I agree is important and but it's also interested that you didn't say anything about Getting good data plane performance. So as I see it one of the key differences between NFV style apps and normal Enterprise apps is the requirements for data very high data plane performance lots and lots of very small packets and you sort of Mentioned SROV and passing but you didn't address the question that today The options for delivering a high data plane performance on Intel hardware are either SROV Which isn't supported by OpenStack today or Something like a DPDK accelerated open V switch, which again isn't really supported by by OpenStack So do you recognize those as issues and what we'll have what you see as the solutions You want to take on this one? Here's our chief architect Identify the situation where the loss of enhancements do need to be added to the incumbent OpenStack communities Technology is complicated. Hello. Yeah, it's okay so Huawei is actually very very actively working together with our telecom operator partners to try to identify the potential Gaps between the the the commercial availability of these OpenStack based standardized the infrastructures the supporting all the various kinds of telecom applications and elements on top of these infrastructure support and we are also contributing series of mentioned the so mentioned Enhancement of these infrastructure layers proposed them to the OpenStack communities You know design panels for discussions and whether it's the these Enhancements both said the management layers of OpenStack as well as the hypervisor layers to be accepted and embraced into the Community open source communities. I think your specific question is that traffic That data plan performance In special regard to data Plain performances we have worked closely also on the DP DK as well as Huawei Proposed the solutions of direct data plane pass through for for the mapping between the data interface cars and processing General x86 based processing cars. So we should call it as a as a net map solution. So these these ones we Besides these there might be also Open-flow based, you know Harbor switch solutions To under the control fee as in controllers of the all the nurture and plug-in Yeah, there might be several various standardized options to support and to propose to the Open source community So I think we're still in the early stage to explore the first we identify what the problem is and And we did we even run a workload on William will platform We see the second degradation on the performance. So we want to figure out Right now we're not proposed which one the right way to do we just figure out what the problem is and Systematic to see whether we can approach it Hi, as you mentioned that's the SDNs could be a long journey and now we are just getting started and in early phase of Taking a look at the technology and how it helps from your typical experience. What is the? Return on investment that's being seen like you know It's a hard dollar cost savings or is a new revenue string that is being created and what is the? Return of investment in this early stage that could be realized through the technology Well, if I guess there's no return at this stage. This is all just investment And but but is the future so you're betting on because there's a workload here and other people doing the SDN Worker really the complexity of network block lots of things happen That's why a Google a lot of what doing where well because they really make their infrastructure fluid Follow flow their traffic and the rest of them are not enterprise is a fixed workload So the driving force actually less powerful in our word talk is actually stronger You just we just couldn't find the white platform So in our practice we do try as they approach but I think it's probably only to say in dollar term how the I Generally as for from our point of views the the most pragmatic near-term Investment Benefits of SDN as majorly the within the data center to construct an overlay network Which will greatly improve the networking administration efficiencies Based on these SDN Technology, this is the most near-term Profits that could be gained Yeah, some of company Right, so for manageability visibility SDM does doing good. So SDM based on monitoring I think based on manage provisioning It's much efficient because they can automate the process Connecting the data more system So that's maybe the only Just going back to your comment you made about five minutes ago about communicating with some of the Current open-stack implementation some of the requirements on gaps Could you give us just a few examples of what they might be, you know I'm curious to know what kinds of requirements and what kinds of gaps you're Identifying and Specifically, I'd like to understand a bit better about your requirements for service chaining, you know, as you start chaining services together I think I didn't mention that one. I think that you you want. Yeah, that's a So the we don't really officially communicate with open-stack, you know, we are part of open-stack. We are we're together so We we identify the gap Other than through the our Experiment we're just actually doing we're not just writing paper and look at what's going on We just move that because we know the app we write the most apps and our partner operate the most apps And so we move the apps and to this platform see what's going on and then we find a place and then we try to solve this problem and I think the service chain You guys probably charming I will have to go to that level yet, but Maybe you guys can add on yes some examples of as you mentioned the proposed Enhanced solutions on open-stack especially novel The examples for example the Federation of multi-data multiple data center Open-stack Previously we we've proposed the solutions of for example a hierarchical scheduling mechanisms of multi-data center open-stack, which is still a blank areas for open-stack and then that Finally is the within the internal discussions. They found that it can be converged into the Federation scenarios This is one case since the other is the Affinity which is the multi-factor based and not only single factor single compute storage and networking factor based scheduling guidance, but also considering the correlations limitations between the virtual machines so the by taking these limitations or confinement into considerations during the scheduling Mechanisms this is another pro pro such kind of a proposal just Within the topics in the in the design panels. Yes, which is ongoing these days All right a question for Alex if you will Can you just sketch briefly some of the you know the NFV or SDN projects at Vodafone has a got going? That's a big question So we've had a successful demo of voice over edity in Germany. So that's running on NFV today That's one of the main Use case that we have today But then you know later on we are we're exploring with the community what the next applications will be but at least in in our German opcode We are really strong on getting voice over edity on NFV so that we can then And make it more manageable upgrade later because we we feel that new features will come with voice over edity And that you know we need to be ready for this Okay, and just sort of I can also Refrain the question asked earlier about you know, where do you see the benefits? I mean in a service like that? Is it is it like the savings the flexibility or agility? Agility may need the agility Since you mentioned about the overlay solution So you think that one was more like the temporary or you still believe in the long run for the telco or For the Huawei for the vendor provider for the open flow disruptive It's more the future way to go because it's kind of a lot of investment to put on this one Everybody's tried to calculate this as SDN is a sexy word, but we don't know how this alright return We know the agility is good But how do we invest on that one any like the early-stage open-flow adoption or the experience you can share with everyone. Thank you so basically our understanding about the SDN's Beneficial to the total customer is that as you mentioned the agility's Trying to build up a fully physical the networking fabric decoupled logical networking layers Which is a which is highly application dependent and after and service is driven So that you can build up a logical overlay network without much dependencies on the incumbent networking devices this is the major advantages we've been identified at current stages for for for the overlay solutions But while the same time we also clearly aware that there's some performance bottleneck problems encountered by the pure software-based solutions we we are Doing these enhancement and curing in two directions one is the hardware accelerations for example to enable some Switch boxes tour switches boxes or some new cars to be able to off-loading these processing while at the same time of course they to Optimize further on the x86 platform based performances for the user plane One follow-on question One follow-on question on this one is if you have the hardware tour switch already can do the overlay So that one kind of improved the performance Why you still need the open flow or the data plan and control plan separation try to do the work Yes, the we's the if the tour switches is acceptable for customers then maybe in some cases for for example our word from customers they they might prefer a Cost hardware based solutions in that case then maybe we switch based Standardized overlay solutions might be a better choice But for some some cases if the hardware and software hybrid solutions Accepted then we can use the ladder and especially We in the multi data center consolidation scenarios we might normally encountered with multi vendor a network Solution providers in areas where we're the sink that the integration clouds solution provider might not have the dominant power to to to urge the all the engaged networking providers to to provide the compliant solutions According to the integration provider Requirement yeah, so the overlay in this case is the best approach to solve this multi-vendor compatibility problems, okay, so I have to