 Good morning everybody. Thank you for coming. Welcome to the Cisco sponsor track sessions here at the open infrastructure summit. It's a little early so I'm slurring my speech just a little bit. Thank you all for coming. This is the second of four sessions that we're going to be having today. We just had a little session on a very high-level view of the telco stack. We're going to go next with Ian Wells and Chandor Ganguly from our cloud platform and solutions group and we're going to go into virtualized mobile networking using OpenStack and I will turn it over to the trustee hands of Ian and we're off to the races. Good morning everybody. So what we're going to talk about here today is I get to give you a little bit of our philosophy about why we design things the way we do and Chandor will give you more detail about what we've learned as we've built mobile networks using as a company obviously Cisco sell quite a lot of components in this space. We work on the platform element of this which involves OpenStack and the needs of a platform in a service provider how to how to deal with it how to operate it how to tune it and so we'll talk about what we've learned and how we work with platforms as we're designing mobile networks. So as I say the first part of this is sort of our viewpoint how we see this and where we how we tailor our platform to suit the use cases and then you'll see some practical ideas of how that's put together in a network in real life. So the state of the SP segment this is really our viewpoint you can see the curve on the right there might look rather familiar we have a peak and then a trough as people you know expect a great deal from a platform like OpenStack like NFV if we're being honest and then it turns out not to actually meet the hype but as people become into realisation it is actually practically usable and you do get somewhere with it so we were sort of levelling off on that enhanced productivity part of the curve today. So again it was maybe a little bit hard for us in the beginning because we were trying to figure out what we could do with this what was practical how to get this working how where the problems lay honestly with the system but it became mainstream maybe three years ago give or take we're now looking that particularly in mobile networks there are a lot of moving parts in a mobile network where software is actually the best way of delivering the functionality that you're looking for and so being able to come up with a manageable mobile network where you can do upgrades in place where you can keep your customers satisfied and where you can run all of that on a software-based platform is an incredibly useful thing. Another component that's or market that's been taking off for us is managed services where what we're doing is helping service providers deliver a set of managed services a menu of services that their customers can consume for for enterprise space and the initial driver here was definitely definitely on the cost saving aspect of things but I think as things have matured then service providers now realize that there's a great deal of agility that can be gained from doing this the the idea that you can if you start going down the path of a decent CICD model then you can test your bring your software together it's so much easier to bring software together than it is to basically get a box from every vendor concerned into your lab every time you're doing an update and then you can run through automated tests as you do that within your lab in order that you can be certain that when you bring upgrades into production you know that they're going to work without causing you problems so agility has been a very interesting aspect of what works here. 5G obviously we all recognize that 5G is coming the important thing to understand about 5G is it's a sea change in the products and the protocols required in a mobile network so again it's one of those lift and shift things that goes on roughly every 10 years and it means that there's a big change that's coming to every mobile operator at this point in time so that means new deliveries of different functionality which tend again to come as software in these days and edge computing is obviously a very exciting area for us right now the idea that as we as we get to the point of having software deliverables we can actually deploy them once or many times and if it should not be any harder to deploy them a hundred times than once or a thousand times so you can now push the edge of your network further and further away from the center and out in towards your customer and that can deliver a number of bits of value so we have progress in terms of getting the performance that we're looking for in these applications making sure we can run them as fast as we can get the maximum value out of a server also in standardization trying to work out what the standards tell you versus what actually works in practice has been one of those journeys that we've all had to travel and the early deployments particularly in the edge side of things as we start to do virtualized ran virtualized radio networks has been again an education for us and there are challenges obviously within a service provider you've got an operations team and now they're operating in an entirely new and different world because they've got platforms and software to deal with so kind of developing a culture change within service provider organizations is certainly one of the hurdles to getting full implementation of this because again many moving parts here that you're getting visibility you understand when something is going wrong as soon as it's going wrong and what you should be doing about it trying to get those processes in place and trying to get the observability in place to make that possible has been a learning curve as well and managing those expectations making sure that everybody knows what they're getting but also this is as far as you go and no further you can wish for more than this but at this point in time this is what works that's been a learning curve with our customers and it does all come down to money the reason that this is driving being driven forward is because people see that they can either deliver more value to their customers they can make more money out of their customers because they can offer customer experiences that the customers can never previously have they can improve on how they use their network also they can get services out there faster because again that agility means that you can do more faster and you can you can get the results to your customers as soon as possible but similarly money it has the flip side which is how can I save money by doing this you can make more efficient use of network resources and the orchestration conversation we talked about in the previous session orchestrating not only the virtual part of the network but the physical part of the network getting them to work to a common end is something that allows you to deliver more value out of your fixed assets your wiring in your service provider network OPEC spend on operations again Chandra will join in with our solution part of this later on but basically the art with delivering software is to make sure that when you're running a hundred or a thousand deployments it is not costing you a hundred or a thousand times running one deployment so keeping the operational spend under control is absolutely key to making this a success so the deployment strategies here we've drawn up that we've seen all three of these right number one is where people were hoping this would go in the early days basically mix and match you would pull the different components from the best vendor for the task the cheapest hardware a virtualization stack that suited you or maybe multiple virtualization stacks because they had different strengths and weaknesses and then VNFs from whoever and orchestration from whoever as well but there are a number of limitations with this one of which is that the mix and match means you've got a lot of touch points between these components components and a lot of integration tests to do to actually find out if this is going to work also mixing and matching here components that ultimately don't need to be different leads you to the fact that you've got a bunch of mops of operations procedures that you've effectively got multiple copies of simply because dealing with one plate virtualization platform dealing with another virtualization platform how you do that is different number two is very much favorite among vendors I would like to say we would love to deliver you everything top to bottom built by us integrated by us it does work there's no two ways about it but it's not practical because again you end up with a bunch of operational procedures you ultimately end up with silos and it can actually lead to the service provider organization itself becoming siloed so I own this this is mine I will not share my experiences or I cannot share my experiences a common platform and a common Mano stack has what have been what we found works best with the customers that we've been working for it leads to a platform which is best of breed which is capable of running VNFs from whoever you wish to buy from and the Mano stack should be common because you're trying to orchestrate all of your VNFs to one common purpose to deliver a network this one is not perfect again there's no perfect here it does involve getting the service provider to actually see the benefits involved outside of their own domain so you're talking to transport people you're talking to mobile VNF people perhaps the difference between voice and and data in a mobile network to take an example and trying to get them to see that the success of their company lies beyond their own perspective but it's the one that brings the most value if you can get it to work and absolutely the majority of the market today exists in models two and three model one is still a dream but I think it's a dream that we have to evolve towards rather than one that we're going to jump to straight away so this is a fairly I'm getting very tired of drawing this diagram this is a fairly simplistic representation of a service provider network and it's you know we like to talk about edge computing this is got a core and it's got an edge in the core on the right hand side there are a very small number of sites involved they're very big sites they're very well managed sites they've got people in there actually you know conventional data centers with decent cooling and good quantities of power and good quantities of backup and there are plenty of things that you can run in those cores you can run a mobile packet core you can run the VNFs that control your network that determine policy and billing for your network you can run the elements that actually manage your network the OSS elements that deal with whether your network is operating at peak capacity whether it's tuning you want to do to it observation logging and so on you can run things that we normally run at peering points like third-party CDNs NAT gateways to make your v4 addresses stretch further you can run video transcoding as well to ensure that your your video is delivered in the most optimal way to the various devices on your network and that works just great but if you look at what's on the right hand side then we start getting to further and further out to the side of the network you get less control over the sites the light sites are less suitable but the sites are a whole bunch closer to your customer now obviously at the moment today we can deliver services on the customer premise with a box that that's put out there but what we'd like to do is work out what we can do with these services with these sites further and further out and that's been the journey in recent times figuring out how to get the additional value on that side of the graph and you get much closer to the customer and that can lead to a couple of benefits one of which is more experience that you can deliver to the customer one of which is that you can reduce the amount of backhaul on the wide area network that reduces or increases the value you can deliver with the same piece of fiber and there's elements that we're seeing that come out here now so packet core in a network slicing world in 5G can be on-prem for instance a virtualized RAM has limitations to how close it's got to be to the mobile radio heads so I need to be within 10 kilometers of my antenna to virtualize the radio access network part of a 5G network and going further and further in then we start wanting to deliver additional added value services that we can do because we're geographically close to our customer so the key considerations here are trying to figure out the cloud infrastructure you're going to do whether it's going to be virtual machines or containers what's cost effective when you multiply up a thousand times over is not the same as what's cost effective when you put it in the core the performance is an IT level of performance is not necessarily going to deliver a packet forwarding application they have different requirements and ran in particular is also a very real time application which can lead to problems in virtualization deployment network services similarly this is both about the control of the VNFs that you're dealing with and the control of your one that would you want it to be centralized which can lead to connectivity problems and can lead to overload as you basically condense everything into a single location distributed which can leave you with questions of how you're going to manage that VNFs the same am I trying to put the control plane in the center and the user plane out at the edge where we're forwarding traffic trying to deal with service chains which are now more often than not into data center they're not simply in a single location being ready for the 5G and the challenges of network slicing that come along with that and again I would focus on the operations model because the expense of operating this network can't be underestimated if you get it wrong it will drain your pockets so understanding what you're going to do in terms of life cycle management of the platform and the applications of operation understanding when you have a problem and how to fix it without having all hands on deck so our approach to the service provider segment is obviously built into our products at the bottom end of this we have the virtual infrastructure manager the VIM which is a product we sell delivering not just open stack but the manageability that's necessary to make that work in a service provider network at the top end the two parts of the Mano stack that actually orchestrate VNFs we have no religion on whose VNFs we're running we obviously supply some of our own but we really expect any service provider would choose the best VNF for the task from the appropriate manufacturer so it's not going to all be Cisco in that in that location and so we are evolving from the left to the right we certainly started with vertical stack solutions we are I would say most of the way to the right at this point in time we work with our own hardware we also work with third-party hardware we work with our own VNFs and third-party VNFs we work with our own Mano and third-party Mano and we have examples of all of those today and now I'm going to hand over to Chandru who's going to talk about the product itself sure so thank you and so so far what you had shown is really the philosophy in which we have evolved this platform what I'm going to talk about is more practical implementation of this platform and where we have installed it in scale so and some of the typical use cases and challenges that we have gone through so what what I want to point out here is typically people in this conference on and lot and similar conferences really focus on open stack or Kubernetes or whatever the underlying infrastructure of the platform is what we are trying to show here is that's definitely the anchored and one of the most important things but if you but along with that you also need to consider other day to you know factors or operational situations that you will encounter because that is where once you have the cloud up and running your costs last tissue TCO optimization will come through so I'm not going to go through every details of it but I'll just take let's say one particular factor let's talk about the life cycle manager there we are talking about let's say your cloud is running a hundred node cloud is running and one of the nodes go bad how do you go ahead go about replacing that node without any impact to the rest of the VNM standing on the cloud similarly let's say there is a security bug that came in and you have to do a software update how do you go about doing a software update and many of the security bugs include a reboot of the compute nodes how do you go about doing that without impacting traffic so those are the considerations that I would request that when as we are choosing a VIM platform thing through those and that is something that Cisco VIM has been incorporated in many of its you know as in this journey that we've gone through just like also another thing is just like you have an API for open stack we also have an API for this cloud platform to manage the cloud platform so the advantage with that is now if I have thousands of clouds I still have a common API and I can manage all this clouds from a central location and that is quite powerful you don't have to go to you know SSH to every cloud or the management node of the cloud to do anything you can manage it all centrally from your OSSBSS system and and typically a cloud platform that you see in the industry doesn't have it but we have that is something we have considered and that entire REST API based cloud management is over TLS and secured and obviously you know potential for bringing in certificates and whatnot so how how has this evolved so obviously based on customer feedback what we have what we initially started with is the CDC where we have a full-on stack like control compute storage all separated out all you know dedicated nodes associated to that what we what we heard from our customers is based on the location you might end up in a situation where you might only have a very small physical location right so you might have to collapse the cloud into an extremely small footprint called MicroPod or as you go towards the very edge you can you actually don't even have space to put storage so we have come something called an edge pod that we have developed and I'll talk about a bit more about the edge pod a bit later and now what we are doing is we're actually expanding it to what is called nanopods this is the work that we are currently doing as you can see now a given customer and I'll talk about the customer in the bird has a has a footprint based on the location of various form factors of the cloud now how do you manage this cloud this is where a common API which will help manage doesn't matter the cloud type or pod type you can actually manage it from a central OSSB system through a rest API is what we are talking about obviously on top of that is your you know vnf manager or you know nfvo it can be again one or multiple but obviously the idea of typically a customer would be to have one so that they don't have to retrain their guys to have three different nfvo's if you will so this is this is something that is existing today in Savim and we'll talk about use cases associated to that so before I go to talk about use cases this is a laundry list and this is not only a starting laundry list of things one would I would request that if you are considering cloud you can you know plan plan it out as part of the evolution in your journey because these are the things operational challenges that come into play on day two day three once you have the cloud up and running for four or five months now you have to go do a software upgrade or upgrade what you do so this is this is kind of a laundry list of what it is and any one of these needs to be factored in so let's talk about two case studies I'm going to talk about so these are practical physical implementation that is going on and actually that has happened in the world right now so it's not like anything is theoretical here the first one I will talk about is what we have with Waterford India Waterford India I would say is a brownfield so they already had a Cisco mobility they were a Cisco mobility shop running ASR 5500 they went wanted to go from you know those physical boxes to a virtual world and they obviously bought Savim and and on top of them they were running the in Cisco's virtual you know packet code but what and this is currently running in in 13 cities in India so let's say if you are in city in Guwahati or Mumbai or Calcutta you actually are running on a virtual network or for you if you are on Waterford India so there are 13 sites that already are in production and they are expanding that to 40 sites and in the 40 sites the one in like Mumbai and Calcutta are huge like because of the subscriber like coaching is typically about two million subscribers so those are pretty large but at the other 40 sites because they are small towns they I don't have they don't need to put in a full pot they are putting what is called a micro pod with additional computes because they are the subscriber scale is much lower but what they have done with this now they have got it to a point that right from racking stacking powered on to getting these cloud on to getting the VNF on in three days 72 hours they can now do it they are now got it to the point that it's pretty much a cookie cutter operations model so this is the agility that they have got also there what they have done is they have updated the clouds in more you know this 13 sites I've talked about they started in January of last year not this year January of last year and over this one year these 13 sites have gone through software updates as well to take care of security fixes additional features we also support different combination of hardware so they wanted the latest generation hardware all of this has happened with software update without outage without any outage so data plane continues to work we have a mechanism to do the software update wherever reboot is needed we deferred the reboot and operated decides at what point a reboot is needed if they will go ahead and do the reboot this way what happens is data traffic continues to flow so that's one case second by the way this is all open open knowledge it's the press releases below right there second one is what I call is the latest one is the racotan model which basically racotan is a greenfield use case where they never were in the virtual you know virtual mobile world they essentially wanted to bring in and get into the mobile workspace what they did is this so they have no legacy hardware it's all virtual all right from day one but where they differed from most of the other service providers is they actually took the more virtual world to the edge so that the videos and vcs are the so the ran radio access network is also virtualized which is the first in the world and there are quite a bit of challenges and I'll talk about a bit on what those challenges we went through this is an year in one year in progress and as of today they have 700 clouds already up and running in Japan in you know in and then they have actual subscriber 5,000 or subscribers already accessing that network what it what is important also this is a pure IPv6 network so there's no IPv4 across in any packet that is exiting any of these clouds all IPv4 IPv6 so what they have is in the in the center they have a CDC and they have at the very edge what is called GCs or VDAN H clouds they are also evolving to what is called RDC regional data centers so as they scale up what they have is a few a handful of clouds at the core a much bigger set of clouds at the at this what is called regional data center and a lots of small clouds at the edge now obviously this brings the numbers we are talking about eventually will be in thousands so today there is hundreds and what so to evolve this what they did is they actually came up with a very very innovative way of and deviated from the traditional approach to obviously they want to make sure the user experience when they go from any part of Japan even to the as they're evolving to the roaming is the same they're also evolving from the legacy architecture to the distributed cloud native architecture so with the CVM also supports and one of my colleagues will talk about the CCP or Cisco continue platform top of CVM one of the talks later we'll talk about that obviously we've they've they've taken a model of going virtualizing the rant and one of the things you will see a lot of service providers do multiple SKUs they have standardized on a handful of SKUs why because they have to operate this huge number of cloud just scale they have their own OSS PSS system they wanted every customer to provide rest API for everything so that's something they've done so they've normalized on that they don't want anybody's UIs because at the end of the day so many vendors are coming in the number of UIs are just insane so they said everybody give me a rest API and I will write the whole UI myself that's what they went with the whole cloud operation actually is today done by actually about less than 100 hundred people today you know and completely CICD based model every vendor that is participating all has what is called you know is a participant in the CICD model so if you look at it what they have done is they've come up with a common hardware common VIM all the NF VNFs can be any one or N number of vendors and with a common Mano stack actually to be truth be told they have two Mano stacks because of time to market that too but essentially their plan is to converse to one the why this works is basically everything has a common rest API so you can have multiple clouds but on the top the same API so it doesn't matter whether it's a micro port or an H port or a full port they all work in through the same API this is a high chart of all the various VNF vendors that is hosted by the CVM cloud the total today of about 189 VNFs operate in this you know in this model so and the vendors anywhere from Nokia to Altios start to Cisco to in Hawaii to Mavinier all of that is floating around here all some of those are written here so one of the challenges that Rakutan came up is this cloud at the edge where there was constraints on hardware constraints on power constraints on space so what we did is we actually came up with a modified CVM to essentially maximize obviously fully automated autonomous self-contained cloud but we maximize the CPU so what we did is collapse the control and the compute control and the compute on the same nodes we only use like a few cores for the host rest we give it to the computer for the workload we also essentially tuned the platform to the kernel that is running as a real time kernel so that way you know because of the VLAN latency requirements you could not put a normal you know regular OS operating system but at the same time we made sure that the operation experience is consistent whether it's the H port or the full port because the same REST API we also made sure the entire all the clouds we are talking about has full monitoring integrated in it's part of the cloud as an option obviously we we to for to run VLAN we also added hardware acceleration on to the same cloud so in conclusions what we will what our main goal here in this talk is yes CVM can bring in a complete and proven solution to meet your requirements obviously the adaptations we have done we have to make this cloud open that is it can run on multiple vendor's platform if you will the difference queues in the case of Rakuten it's running on quanta hardware in the case of whatever one and you're running a Cisco UCS we also support other third-party where platforms as well we've also made sure that whatever we do net TCO at the end is lowered so you as you scale out it's not like you need 10 more people as we've got cut into scale out it has to taper down your operation cost cannot keep on going up that is the most important as we have designed this we've also made sure it's secured the cloud can be monitored for both from blogging and assurance point of view also we proven which is kind of the first in the world that VLAN workloads can be can be you know put on a cloud so we do you know thanks to many of our customer we have got some experience in this journey but as the next step from this is wall being to CNS or container platforms one of my colleagues in to the third or last talk of the session you'll talk about how he's using that you know he will give an example of how Cisco container platform is also running on CV with that I conclude my talk any questions a couple a couple of points before I go into some time for Q&A with Chandrani and all these sessions all the sessions that you come to at the summit are recorded obviously and these all go up very very quickly on the open stack foundation YouTube channel so I know a lot of people are getting pictures of the session as the session goes on but full recordings of all the sessions that you come to here at the summit will be up on the open stack foundation YouTube channel so you can go back and go through any of the details in the session the other thing I wanted to mention was again as Chandra mentioned come down to the Cisco booth in the marketplace we have demos available on all the technology that we've been showing here today especially the virtual infrastructure manager for the CVM product and with that I'll open up the floor to any questions for Chandrani and wow you have successfully explained everything it takes to virtualize a mobile network I'm very impressed I'm not actually more impressed with you as I am with them eyes because they've already got it figured out I let the team know you get a raise any questions at all for Chandra and Ian or we can pick up the discussion down in the booth like I said down in the marketplace okay so we expect virtualized mobile networks from all of you by the end of lunch or maybe not okay well thank you all for coming we do have actually the next session that is coming up starting at 1040 1050 at 1050 there's the morning break coming up but at 1050 we are actually very fortunate to have a member of the Rakuten team Ashik Khan who will be here and presenting a deeper dive on what Rakuten is doing with their mobile network so hope you can come back and join us then at 1050 thank you Ian and Chandra