Hyperscale SD-WAN - VMware VMworld 2019
Transcript
hey guys what a Steve Townsend from the CEO of laser comm you know I get to remove my networking rules we're at vmworld 2019 and physically we have with us from the bailout team so I think the whole official title Tyler Craig its VMware SP win by béla file that's correct but more importantly we have a chief architect right that's right credit Connors thanks for joining the program Bello pal SP man SP man has been a big deal for some time you know you is the
the trickiest band to get into the sea fleet and and displace some of the comments it's all been all about coffee etc we're going to go back onion up on that a little bit tell me what's the big kind of news coming out of bailout FB win real world tweet 19 so here at damrell 2019 we're talking about how SP winds becoming hyper scale so SP man you know it's been a big topic for a while but now we're really reaching the point where it's becoming
almost ubiquitous in the industry everybody's talking about it more than 150,000 sites already deployed on VMware SP life so let's talk about this type of scale story a little bit so cuz you do me back when you say es un is hyper Caroline hyper scale by using SP yeah or you know so you're saying Bella pal has basically become it hyper scalar because of the 150,000 deploy 150 sites the boy okay so help me out there $150 a size deployed you know what in comments Cisco
or juniper they have millions of sites deployed what makes bail without a hyper scale versus one of the comments right - you nasty winds a lot different than traditional routing you made there may be millions of Cisco IOS cars out there but someone's going in and managing device by device st wayne we flipped on its head we built this horizontal scalable cloud service centrally managed centrally monitored bringing all the statistics upstream being able to perform analytics in one place and to be able to do that
across 150 thousand sites we think is a really interesting problem that we've had to solve so listen sorry let's start to break down the skill piece might I might buy this ideal of Bill cloud is a hyper scale or if we can help me kind of give some context around the data points let's talk about how much data is coming off one side so you know one site it bears no that varies dramatically based on the size of the sites we have everything from you know
manufacturing sites in China that have 20,000 employees to insurance agents at home that are just one user don't get any breach but but across you know across our 150,000 sites we have tens of billions of flow records coming in now every five minutes so we've got a huge amount of data coming in that were processing insights into how the application is performing how the network's are performing around the world you know based on geography based on applications that users are using really really valuable data for
us to continue to improve our SD land product so I wrote a piece of a couple years ago for Czech Republic where I call SD whammy big data problem you know if I look at what I'm trying to do one friend just the customer last week who was trying to do some policy based reality they wanted to send voice down one half and replicated out another and then flip it kind of manually I'm like guys this is Seoul 2015 you really need to look at SD
win because it can take that data real time in route packets I understand skill that the reason why we couldn't do that before it was because of CPUs and routers and the algorithms just wasn't there memory and all of that evany when solved that on the micro level now you're talking about the macro level and billions of flow records how do you process that much data I mean that's a lot this that's that's lovely so you know there's there's two different angles to it obviously advances
in cloud technology being able to build horizontally scalable cloud services doing ml doing big data analytics that's an important piece of it but just like you would have heard if you saw the NSB Yukino of this year talking about NSX intelligence Tom Gillis talked about not just bringing the data to the analytics engine but also bringing analytics to the data one of unique things about VMware SDRAM is from the beginning we built this hierarchical model we have the edge device just like any other you know
SDRAM done here but we built this network of cloud gateway thousands of five gateways around the world not only do they provide us a distributed control plane in the cloud they build us another layer in a hierarchy so we have analytics happening on the edge where we see every single pack around through we have correlation happening at our gateway or a single device is seeing thousands of edges that are connected and then that data is advanced at upstream so you guys haven't used the buzzwords at
least talk maybe email AI etc but this is that traditional model where we have you know this huge amount of data injection at ant we apply the algorithms at that edge any anomalies get you know brought up into the central control plane where we do additional processing so as we learn stuff at the edge we don't need to send all of that data to a to the cloud for analysis we can send a subset there and then we can perform some additional data mining yeah if
we need more data you have 150,000 s points you can get bored with bigger promise that generally the yeah that's right I mean we haven't talked about in the past we're gonna be talking about it a lot more going forward the big challenge with analytics and that's that it's sort of problem space is that you need the data you're right you can have the best MLM algorithms in the world but if you don't have the data if you do those algorithms it's not going to work
and so we built the network we got the customers we got the data and now we're taking that to the next wave of innovation which we believe is the future so how does that service up to the end-user the customer so you know VMware has all this great data on the back end but what value what insights can you give the customer base long-day information yes so you know customers have already benefited from this whether whether they know it or not under the covers traditionally a
big problem that customers had to solve is configuring QoS my blood pressure just literally my blood pressure just raised what'd you say - last period a minute that just we fix me get angry just like nothin personal against you is just the concept of battle us over mine or my work right so that data that we have for instance right we use something called smart QoS inside though we we try to do QoS automatically and if you look at that customer base that we have deployed
92 percent of our customers are not configuring any QoS policies on their VMware Instagram they're letting us manage the QoS and provide them with a great experience for all their apps so what's the results like our customers leaning into it are they calling you and yelling at you from a support perspective you know what my what looked like FTP traffic is some type of real-time data and you guys are stepping on it what what then would be so I can tell you that the eight percent
of customers that customize their QoS create more support cases for QoS issues than the ninety two percent that do not and rely on our algorithm yeah you should have started the whole conversation with it's like you know what and being where I write this point that you know Bella pile is kind of integrating nicely with being worse overall approach when I deployed vSphere out the box normally I don't need to do anything other than to point it is pretty well tune based on being worst years
of experience so it sounds like bells house continuing that tradition of it you know what you really have to be the exception to to want to turn knobs in this solution right and then we want to extend that beyond QoS right so we look at what other problems our customers calling support with and and how can we build algorithms to detect those problems so that's not the end user calling in saying there's a problem the product is telling you this problem and the best case the
product is fixing the problem for you so will in one less question on just more of an integration question between Bella filed Indian work you know about a pile of dust SP man which is dough man in a sex in SXT is the data center but when we look at what the two products are trying to achieve they look very complimentary so you you know we had the NSB you GM on the program a couple years ago right or even have two years ago right during
the time develop out acquisition we asked the question hey when are we going to get integrated SP man and integrated data center I'm holding their question now for the chief architect work Bell top where we're going to get integrated as data center an integrated SP with so there's obviously points of integration now we're gotten we have realize Network insight giving an overview at one single pane of glass visibility and what's happening in the data center what's happening in the branch we have extended micro segmentation from
the DC all the way through the branch leveraging Bell about nano sects working together but at the same time data center networking and branch networking they're not the same problem they're not the same in the case of most big companies they're not even the same team that's managing it so integration is happening and there's a lot of natural points of integration especially around visibility and analytics and being able to pinpoint where the problem is happening but we're not gonna try and you know jeopardize the strength
of either product line by trying to over integrate so at some point we'll have you on the future program will drive will please the networking geek so though so we'll have like some deep conversation about the value of each VPN and BGP and bringing that stuff into the data center I'm kind of I'm kind of over the EVP in concept but that's my personal opinion well you know what we'll save that for later would be great really appreciate you joining the program you guys want to
find out more join the conversation hashtag CTO dose one Twitter or only then you can you know I'll tag created I'm sure you're only ten you'll get a ton of alerts when people need you like what that question Keith actually was completely unfair or keep us right on point if you want to contribute to the conversation probable hashtag hashtag CTO dos talk to you next CTO DOS
