 Check one two. Good day. So we're um, we're just gonna start by giving some introductions We're actually waiting for our panel moderator to turn up, but in the meantime, we'll Each say who we are what we do I'm Lincoln Dale. I work for a company called Arista Networks and at Arista. I'm a principal engineer Arista is Well, I guess typically the way I like to explain it is one of the most unknown networking companies in the world but probably in More like an open-stack environment. We're probably more well-known. So And our involvement with open-stack is around a bunch of things related to neutron and that's why we're here today Thanks, I'm John Vastal with Pac-net and the token carrier on this group. So hopefully they'll be gentle and We recently announced a Kind of pay-by-to-drink if you will Ban what solution that we're tying back into neutron and to entire open-stack community? So we're hoping to do great things with this group Great, I'm a loot Tucker CTO of cloud computing at Cisco Systems Then an early contributor into neutron help get that separation pulling networking out of compute As you know, open-stack is made up of the set of these services And so one of the real advantages of doing that it's really allowed us to accelerate What we're doing in networking and making an architecture that is easily then extensible into all of the different Vendors that you find up here And so I'm glad to see it because I've always been wanting to have a telco provided bandwidth on demand Solution that we could use to bridge data centers My name is San Chile. I'm a CTO for the carrier business group in Huawei Yes, Huawei possibly, you know, just recently is elected as a gold member for the open stack So we're fully committed for our Infrastructure in the data center and also across data center for the carrier specific requirement So SDN is certainly is part of it Okay, it's beyond the data center and also looking into the network edge how we can leverage this New technology Disruptive technology and help the carrier and also data center to transform Okay Hi, I'm Peter Belanda With VMware in the networking and security business unit so I work in the office of the CTO under Martín Casado and most of you in the open-stack crowd probably know that in At VMware we're working on the network virtualization components. We're heavily involved with the neutron framework and We also at VMware is participating the other components of the stack, but I'm here focused on networking So, you know, I can start with the question baby in lieu of the moderator Yeah, we know no, but we will self moderate So maybe we can ask one interesting question We might just ask is this is about the future of networking But rather than starting there and Nicholas Carr years ago said, you know, IT doesn't matter And I hear actually a lot of people saying networking doesn't matter. I just want my apps up there I'm curious how we might interpret that and how we might respond obviously matters to VMware very much now Networking doesn't matter. Well, of course networking matters because the attendees in this room would say that At the end of the day really what business cares about is is the applications to achieve their goals and Networking is a means for those applications to achieve that. So the networking piece of that is to be able to work with the framework of fast provisioning of applications rapid deployment cycles that are happening and that's the the type of Innovations that I really see coming in networking today, of course as we get to the future that'll continue Yeah, this is very good question what we view that the IT doesn't matter and or it's really Implies the whole business IT used to be in the back end It's more moving towards the business marketing customer engagement the whole life cycle on the business side so it implies the back end support, you know, you can move to the cloud you can move and network does matter and That's why we are here and Everyone all believe the network become more and more significant important Virtualization abstraction, but by the end it's the business. How can we solve the IT transformation? And all we called we are called ICT transformation because not just IT and CT substantial in the transformation as well Yeah, I think at least coming from the carrier side It was always the network the application people didn't care because I didn't have to deal with it It just worked right you or it didn't work whatever the case may be and there wasn't an easy way for the application to communicate or interface directly with the network and And I think what we need to do as a group is figure out ways to Integrate the application directly into the network and allow the applications to have more visibility into what's happening at the edge So the IT guys can take responsibility for the entire delivery of that service Yeah, I think totally the network is important to the applications I mean at the end of the day apps need to talk to one another the network is the plumbing for doing that a lot of the starting point where I think Many people have been in is the network has let them down historically that it's easy to provision of EM is easy to provision an app and you might have a systems team Or apps teams that do that but to provision the resources in the network took Hours days weeks months, whatever to get that provision. So Automation is the key. I don't think people really want who I don't think people really want to be running expect scripts or Awful things for automation. I hope we've moved on from that and it's I think that's the sort of the The the starting point certainly for what we see a lot of with the open stack the automation is probably the most important piece Allowing people to bring up and tear down things and sort of not have to be relying on a on a human manually configuring all those things Yeah Sorry, I was running Scheduling conflict. My name is Mark McLean and the program tactic will lead for open stack networking project neutron. So Oh Perfect Does networking matter absolutely to me. Yeah, so you talked a little bit about the ability to You know have an ability to orchestrate and have the applications configured the network You know, we have some of that support today. We're looking a little bit down the road You know three years from now What are some of those next steps that we can do some of the basic constructs and in terms of orchestration? Yeah, actually, I just gave a whole talk on that But I think that we've seen in when we think of open stack as being this Platform layer between applications and the underlying infrastructure whether it be physical virtual and everything else We've been very intent in the entire neutron project of making that communication happen having an application Express what it needs either at setting up its own logical isolated networks And I think we're we're we're seeing it moving as orchestration becomes much more part of open stack We want to be able to involve the orchestration that's happening within the infrastructure with you orchestration That the applications in fact are doing and make those that conversation happen So I view that you know that having this we're actually gonna be in an out Cisco's making this big announcement application centric infrastructure And it's really about though having these constructs and abstractions such as policy Instead of talking about all of the wiring talk about the policy that you want to have the infrastructure and you know Provide for you from these you know the web front-ends back into your mid middle tier and database and everything else So I think orchestration is one of those things that is really Coming now into open stack and it'll apply into the infrastructure as well And I think we have to take it outside of the data center We there's been a lot of focus on how do we do this within the data center? but you know one of the biggest challenges is especially in Asia, it's a big place and Applications are being used and delivered across the entire The entire region and how do we take those same constructs and the same policies and deploy them not only in Hong Kong But in Singapore and in China and Australia and And then globally across there and the carriers have to come up and step up We've been the bottleneck for many years as to doing this and not provide the flexibility and the means to consume network resources the same way we consume compute resources for example and There needs to be changes in that entire industry to do that. I Want to further emphasize what John just describes. Yes, exactly the distributed data center because the intelligence edge and Edge network edge for the carrier hardly distributed and this is something of you with the significant importance And currently has not been addressed in the open style in New Trump So this is where area while we also you know as a gold member We are going to further emphasize and how we can contribute This is significant important for the orchestration automation across the carrier network and across multiple data centers And this is area we need to build an ecosystem the a new trial is a great platform The flexibility of create innovation and this is where we are going to further looking into it so One of the things I think that I Guess one of the directions will sort of see things going in I think there is some really neat exciting technologies within networking that Sort of really relevant moving forward From let's say a risk to this perspective. We're very bullish on technologies like protocols like VX land for example It allows you to extend a Virtual network outside of a data center It allows you to go outside of a pod You can do it in software. You can do it in each one today. You can do it in hardware and It sort of becomes it's also a way that you get beyond The construct of like a VLAN so suddenly you're no longer constrained in a multi-tenant environment to 4k tenants or things like that There's sort of lots of things of how I think networks will evolve I I'd certainly agree with Lou that applications are important and obviously this goes Got an announcement on that tomorrow, you know, we had an announcement on that yesterday so it's a it's a it's a it's a fun time to be networking this week, but Where where there will be some really interesting Developments and and it would be fascinating to see sort of how it pans out over time. I think one of the most Exciting things about for example, what OpenStack enables is literally the open ecosystem and it's ultimately about giving the end user's choice on How you go about orchestrating things how you go about managing it your choice of controller your choice of technology really do it in hardware or software or whatever and That's something that I think is I Don't think we've actually quite seen the degree of innovation going on in networking for a long long time So I just want to add quickly that I agree with both Lou and your points on Applications and it's particularly policy with the networking Neutron has provided a great platform for innovation above and below the abstraction and we've seen advances in Application delivery self-service all kinds of great stuff that you can get today to some people. It's it's future still But it's really near-term and it's available as Products today on the policy side. This is something where I don't see those abstractions quite available in OpenStack yet, and I Think that it's actually a bigger problem than just networking networking is a very large piece of it But it's something that needs to Both happen inside outside of the data center and also inside and outside of networking across the cloud We just initiated a blueprint actually with with Cisco and IBM Juniper and others really so we can have that discussion and I think we will find it's gonna be more than just the network Because we really are saying it's a much easier way to always have a policy driven you know a declarative way of saying what you want to have happened and Also accommodate the needs that then the real world needs that you have a group which sets that policy You know and now applications should be able to spun up within that policy and within the teams And so that's where I think we're trying to take some of the blueprint work. Okay What do you one of the things that was interesting is anybody in the audience having questions like to ask of our panelists? Yes So the question just to repeat If you when you take a look at open flow and SDN in term Where where are we on the maturity curve as far as technology goes? Well, that depends how you define SDN and I like Well, you could find it as open flow Martin defines it as cool stuff in networking And so, you know, everyone has their own interpretation of it So there are solutions SDN solutions today. There's several products you can buy install test out in your labs Today, so I think it we're very mature on that But at SDN as a term is so broad and it has completely different applicability in the data center versus in the carrier space So depending on your particular needs, I think there's solutions in very in all areas Yes, that's a very much echo what you just described this SDN There's a narrow band narrowly defined SDN region only but which is now I think everyone even the definition of open flow much go beyond what the regional, you know But what I believe from business perspective of every vendor or target the customers has their own strategy Either is a hybrid model either is overlay or it's completely decoupling model and the open pan out in different market segment each one carrier has our vendors have their own strategy to move at the different entry strategy level So SDN as from that perspective still emerging in a much broader sense of SDN Okay, and think or eventually market will pan out and overall we truly believe it's like it's not just networking room It's more a converged infrastructure for the data center across the server storage and networks and Big challenge in the operation side of the IT, you know The network guys wouldn't talk to an IT, you know server guys or a very different organization That transformation will take a much longer time than the technology self in reality Yeah, I think the key word and that is software And we're whether you're calling a software to find infrastructure software to find data centers It's been a lot of terms, but I hear that from the customers all the time the age of having Network engineers log into a router or switch and make changes then across, you know 45 of those things and doing that flawlessly We know we we just can't go down that road instead. We're applying automation and and so now it becomes what are the API's You know, what capabilities do they have what are the orchestration systems you use? And I think that's actually a richer conversation than narrowly around, you know open close a protocol You know, so what we really are talking about is how do we reach this level of maturity in using software to Instead of just to find your infrastructure, but actually make your infrastructure responses Responsive to the needs of the applications and things that are going on whether it be failures or whether it being flows and everything else like that So the software component is that it's what I've seen that big shift in the last couple of years everybody and now it wants a software to find data center or infrastructure so Given the announcement we had yesterday We kind of think it's mature because we just rolled open flow across the whole network and have exposed that actually to the consumer So I think it's mature enough that we can actually expose it to the developers that they can break it for us The the real challenge we have as a carrier It's exactly what you said it's getting out of people used to logging into a router and executing You know Junos or IOS or doing something that they're used to and now we're hiring system engineers to manage our network And from a carrier that's a very scary thing to some regards But it's also finding customers that are able to go and utilize it now that we gave you these tools And we can support them How are you going to go and build on top of that? And that's I think when we can get that level we'll see it become even more mature than it is today Yeah, so A risk that we basically we're doing I guess what's called STN these days before it was called STN So the idea was that we're open extensible You can do anything you want. So, you know you log on a switch It's running Linux. We don't hide that. So we have people running all sorts of things Chef puppet see if engine Whatever for automation they've been doing that we've been doing open stack for a long long time Sorry open flow for a long time as well that the challenge Let's say with open flow on physical hardware switches is you're constrained by the size of the forwarding tables It's great that you can do these 12 to pull or 10 to pull for matching traffic in open flow The challenges is his finite resource called a T cam that Isn't an infinite resource and in fact on much Cost effective silicon is a fairly scarce resource. So there is I guess Choices for STN You know people who have certainly built applications on it. You can use it for exception routing things like that But you know, are you actually going to be doing holding doing flow based switching of every? UDP flow every TCP flow. No that that just will not scale But you could be using for aggregate traffic doing polices things like that, but At a sort of a high-level STN Tender thing this TN is stands for still don't know It's a you know, you could we could ask everyone in this room today What STN means to them and I'm sure that we get as many answers as there are people here. So but to some extent it's a We kind of categorize it as Allowing you the flexibility to do whatever it is you want to do from the equipment you have and Maybe that you haven't been able to do in the past Yes Okay, so I'll start when we go around this way I guess So the concept of overlay networks, Arista. We were a big big believer in the X-Land and we I think We're certainly one of the switch vendors that has a hardware VXLand gateway capability We're a big believer in the separation of the overlay from the underlay the overlay provides the The concept of virtual tenants V&Is is what it's called in VXLand and the underlay You kind of have flexibility how you build a network at layer two layer three What routing protocols you use? The key of any of it is to my mind is you build a Robust underlay in whatever way is possible. I don't believe in sort of reinventing Protocols for the yeah for the fun of it kind of thing BGP or OSPF works particularly well today, you know, we have Speaking personally we have some of the largest cloud networks in existence with you know pods of 200,000 plus servers inside a single pod and You know, so clearly it scales They have sub-second Convergence, so it's um, you know why reinvent that and certainly BGP is you know used across the internet It works perfectly fine there for connecting people together. So That's exactly it now from an overlay VXLand is one choice. There's MVGRE from Microsoft is another choice Nice era had STT It's a bit harder to do STT in a hardware switch But you know, I don't know that there's a clear choice winner But I think there's certainly a lot of momentum behind VXLand as the underlying technology, but to be honest It's depending on what the silicon supports Yeah, okay, so I Think actually the overlay networks are actually dynamic because essentially you have a control of pushing the policy The explain started out just talking about being explained specifically it started out life with multicast as a transport And that's what the ITF draft for it has the reality is what's actually been implemented by anyone is Or where it's going is to a unicast transport which requires a controller to define the endpoints for where what's called the VTEPS are However dynamic that is is it's the control that pushes out that policy So it could be as dynamic as you want it to be so Not sure I have much to add to that From a carrier perspective we look at outside the data center and we use to me the best protocol needs to be used in the Right environment, and I think there's different protocols that we need to use depending on latency and all the other things that we see in the network I'm an old IP engineer still a fan of going back to my VGP and OSPF days So I think you have to look at where you're serving that and what you're doing in that particular realm and pick the best solution for That we're leaning towards VXLand Just because it's easy to implement and there's a lot of implementations out there So I'll take a stab when it comes to the actual overlay protocols I think we have a variety of options there and that's great if you're running hardware asics VXLand It's an excellent protocol for that if you're running on today's x86 hardware STT has optimizations that make that faster at VMware our solution is hinged upon a network virtualization and what we take that to mean is that we have to faithfully reproduce all of networking in the overlay effectively So that doesn't that's not just limited to L2 But also L3 L4 through 7 services and of course if I say L3 that includes Static and dynamic routing so your off-ramp on an off-ramp into the overlay is a critical component of the overall system and a key area for development So I guess kind of to rephrase with the commodization of hardware networking Standardization tends to run slower. So how do we innovate? How do we keep the pace of innovation high because you know Considering like v6 the standard was from 1995. You take a look at penetration today You know software to find networking like you said, it's kind of hard to now be able to find So we have brand we have different interfaces. Is that a problem towards innovation? I mean John for instance, you've rolled it out Or was that did that enable you to move faster or would you run an unexpected problems? And providing it to your customers. I would say that was actually what allowed us to do it There were so many choices and most importantly is DS and SDN it was software defined So if it didn't work, we could rewrite it or we can route it right around it and we can fix what we needed to fix at that point Without that ability it would have been almost impossible for us to roll out what we've rolled out over the last couple of months actually, you know Historically if you think back in the early days of networking or the internet, you know it was software running on general-purpose computers and So we've come sort of full circle. I think and that we but what we recognize in the time being was that actually the General-purpose computers could not handle the performance. So I think that just like we're seeing GPUs coming into Compute, you know computers to increase their performance, you know better nicks and things like that that they'll always be this sort of Oscillation I think between how much where we understand something really well We can build ASICs and to perform those very specific functions and what we're recognizing now I think is actually it's always software and hardware working together applied where where you need to and so I don't think this is by Any means leads to the ultimate complete commoditization of Hardware because that would recognize that we know completely what we're doing in this and instead it's really the involvement of software with Hardware working together to deliver the network to the future It's a good question. So everyone's in mind in that kind of thinking So I think if we look at from business perspective is the TCO Okay, it's the end-to-end solution So if you can say yes, how we're commoditized the open stack commoditize the infrastructure software That's the end open source commoditize it So if you look at open end-to-end solution perspective from IT perspective, the RPEX part is 80% The hardware is 20% Okay, and the RPEX is continue to grow what make it difficult for IT infrastructure Business transformation is in the RPEX orchestration over cost. Yes, if you look at the statistics show Look at over the last ten years. You look at the server technology price change and Storage network very much follow the Morris law not just the server the chip technology has been happening as well So they will be continually important the certain significant Intelligence in the hardware, especially the data plan the data plan doesn't mean, you know It's dumb. It's a data plan also complicated And it will be continued to be there It's how do you play out the intelligence and the control and how are the best combination of the two and to reduce the total cost of Ownership for and also enable the care are in the operators to transform their business Yeah, I'm the I really agree with what you're saying on the total cost of ownership I think commoditization really simplifies the problem beyond What the reality on the ground? I think when we took look at something like neutron where we have a well-defined platform What that enables is for innovation to occur both above and below the platform and they can occur separately And when that happens you create competition among the vendors of The technology and competition leads to innovation. It also might lead to commoditization in some instances But innovation that competition is and innovation is ultimately the winner is the consumer So I think we're in a really good spot right now with these Interfaces that have been defined So I do have a little bit of a follow-up question. I'm self-interested in being a new term in PTL So we define these interfaces Do you find that the community process is fast enough for your speed slow enough it too slow for? moving It's always too slow You know for those of us contributing I think it's always moving too slow and but for the people who try to consume it It's moving too fast. So we're trying to strike that right balance And we're also trying to do that through the separation of concerns And that's why the way that that neutron was constructed that we can have a slow moving interface for our users That can incrementally add features and we can open up the bottom of neutron or whatever it to for all the new technologies that are being Developed without disrupting what the user expects So I think we've done a great job in terms of the overall architecture allowing those two things to happen that innovation above and below the platform Any other questions one last question Yes, so the question is The software capabilities give developers a lot of power to configure and deploy their application But at the same time you get a lot of when you have that power. There's a lot of great responsibility How do you limit it? How do you ensure that they have success and configure it in a manner which makes applications successful? Good actually, I'll make a very short answer. I think that is a very key question What's interesting? We're we're doing in a multi-tenant world, right? And that's one one that arbitration between the desires of the different tenets is what what open stack has to sort of provide You can't let one one tenant take over all of the Virtual, you know VMs that you can possibly have nor could it take over all these swamp You know pull up all of the available bandwidth in the system. I think that you know networking actually has always been multi-tenant I mean it's been carrying traffic from all the different applications and we've got quality of service and those kinds of ways To try to prioritize traffic, but I think there's gonna be a lot more work needed there As we see a lot more demands being pushed by these applications that now all of a sudden have the power to make these kinds of requests Do we have any more time? Okay, so I'll try to be quick. I think you're touching upon a key point Which is that we do now have all this power to do things, but there's governance compliance and business policy issues that are at play here and having a Foundation for declaring that policy in a standard way that these other components can ingest is I think a key area for development to help tackle that problem Just kind of wrap up as you can see there's a lot of exciting Action going on in terms of networking and software and applications, and so I'm very thankful to have our entire panel here today So thank you guys