Scale Computing Brings Order To Edge Computing Sprawl

hello everybody i'm c.s caraval from zk research and welcome to another uh z cast episode i'm here today with craig theriak vp of product manager from scale computing uh before we get into it though with craig i do want to just give a quick shout out to e-week uh e-week's uh zk research media sponsor uh all z casts are done in conjunction with the eb keyspeaks program uh now craig you're the company called scale computing and then we're going to be talking edge computing so before we get into what uh what you do and uh you know some of the new products you have uh just give us a quick intro of yourself and what scale computing scale computing does sure thing so i'm i'm craig teriak i'm the vice president of product management at scale computing i have been with company in this june it will have been 12 years which is hard to believe um kind of came up with a company when we were a storage company back in the day when we first joined uh made a transition into the the hci space um probably 10 years ago now with the introduction of hypercore and then we have recently transitioned the business again to focus more on edge computing and so that's that's kind of who i am and how i i came to be in this position at scale the company itself as i said is uh primarily we position an hci product where we're taking you know servers storage virtualization combining it together into an appliance form factor and servicing markets that have typically minimal or no i.t resources and that can be in the smb space where it's you know one to five it administrators who are asked to take on just the the infrastructure alongside everything else they have to do uh or in an edge computing environment which is typically a very distributed environment that may or may not have it resources on site there okay so let's start edge we're talking about the definition of edge it seems like everybody's an edge computing vendor today and i hear uh you know the iot edge celluloids there's all these different definitions of edge so from your perspective define what edge computing is yeah so i mean edge computing it's simply anywhere you're running mission critical workloads outside of the four walls of the data center or cloud that's in my mind that is edge computing uh in our case we're selling the infrastructure where you can run that those applications uh and it's you know it's it's i think commonly it's referred to as you know anywhere you know our definitions are on that mission critical workload uh because i think that's where we provide values you know basically preventing downtime and that sort of thing in those environments but but it's really around you know the edge of the network near the things and people that are generating and interacting with the data that that is that that's very consistent with my definition because perspective we define edges is really anything that's not a centralized compute source uh that being you know public cloud or private cloud and everything else is edge which is interesting because that is about as broad a definition of convenience there is right i mean you're talking anything between your private data center and your your public cloud so there's quite a bit in there um but so where are we in this journey the edge computing if you were to use the baseball game analogy or the picture was brought up first inning the relievers coming at in or second inning i mean we are it's it's pretty nascent market um i mean even the term itself is fairly new it's probably you know peak of the hype cycle right now on the term um the reality is that well i guess depending on how you define it if you include things like robo robot's been around for years you know the idea that you want to run computing close to the people that are interacting with it that is not a new concept i think what's new with edge computing is there's massive data being generated from you know iot devices you gave us a good example earlier uh or you know workloads that require really really low latency to make uh you know decisions on the factory floor that sort of thing where the latency of going to and from the cloud is just too much or you know a lot of cases and customers that i talk to one of the biggest drivers of needing edge computing is that they just cannot consistently rely on their internet connection and so due to that it's they need some level of autonomy some resiliency in the infrastructure they can stay up and operational um i would say regulation is another big driver uh depending on what industry you're in or how you interpret the the laws in the region that you're in there could be things like personally identifiable information that you would need to keep locally and then i would say that that last driver there is cost and none of those things have necessarily changed although cost is coming down i would say and there are new interesting ways to solve some of these problems around you know the amount of data being generated and the amount the load latency requirements and all that sort of thing but uh i would say the need for something running locally been around forever edge computing as a concept is in the last few years really taken off and continuing to uh to grow at a pretty astonishing rate and i think we're hitting somewhat of a perfect storm here for the growth of edge because i think as you mentioned we've always had a need to have um you know some types of veg computer in fact there's lots of industries oil and gas for instance where you kind of had to have some kind of edge node because it was difficult to do things locally but i think if you look at this big desire to change you know fan experience or patient experience or customer experience or even employee experience we need to move the data and the workloads closer to the user and to me that's what's driving um you know edge computer and i think in addition to that you know we've had the rise of you know higher performance gpus and dpus we've got flash storage that's available a lot more or less expensive and so all those things are kind of coming together that's allowing the edge competing market to finally take off right and so uh you know i think from uh just you're curious you know what kind of use cases are you seeing um that's really driving edge today yeah it's i mean if this is a broad term it's like every single industry we talk to has something that that is edgy about it uh but the ones that i think uh like we're scale computing yeah yeah sorry about that yeah it's like people always use scale puns around us like oh no pun intended it's like yeah you can't avoid it um anyway so the use cases that we see at scale computing are around retail is a really big industry for us you know might be running a point of sale system customer loyalty program um again it's really when i keep going back to that term mission critical and what i mean is that if if whatever that application is whether it's container-based application or vm doesn't matter whatever that application is if down time to that application means that you're unable to process transactions in case of retail or it's going to have some you know big impact to uh the the ability to get widgets out the factory floor that sort of thing that that's really the key to it for us um and in the case of retail that is the point of sale system typically or the customer loyalty program where if that goes down for you know even as few as five minutes or so you'll see people start to abandon carts that's gonna hit the bottom line uh of course those are running next to things like um you know the grocery retail environment we've got a couple talk through some explicit customer use cases later if you want to but we have a couple customers that are running things like um if you're walking to a grocery and you see the the misters going off over the vegetables those sorts of applications just you know small container-based applications where if they went down probably not a huge deal but it's running alongside these other applications where mission critical is the key um and those are you know that's usually what we'll see in retail and i would say the biggest driver there is the autonomy right so it's like you've got you might have several locations you know right in the main part of town where you've got a great internet connection you might have somehow you know middle of nowhere areas of the world that just don't necessarily have a great internet connection so they need some level of autonomy to run their infrastructure on chrome um maritime is another one that i think is you know talk about edgy it's about as edgy as it gets right these are these are seafaring vessels going in and out of international waters and probably in and out of cell coverage with with limited bandwidth and ability to actually get data to and from the ship uh and so some level of autonomy is required to run their accounting systems cargo management that sort of thing is interesting because lack of solid coverage means you need edge but also 5g allows you to put edge computing in places you couldn't before because now you can get data to it so so that's going to be but let's let's transition to scale computing itself you know you mentioned you've been there about a dozen years when i think of scale i was introduced to you a few years ago as an hci vendor in fact before that i think people people that are familiar with you might know is you uh you know a robo you know hci vendor so what have you been up to in the edge computing space can you talk about your portfolio there and how you make edge competing deployments easier yeah um i appreciate the the refocus a little bit so so yeah if you've heard of us you've probably heard of our hdf product uh it's called hypercore uh hyper core operating systems the software that runs on the nodes allows you to run these these clusters on premises uh what we've developed in the last few years is um i've been working on what we call fleet manager that we're happy to announce actually this week those two things combine into a really comprehensive solution that we call sd platform that is explicitly intended to be used in these edge computing environments um and so that's um that's really what we've been up to i would say one of the interesting things about our hci play is that we those of you that really know us understand that we you know we're bringing the virtualization layer alongside our old uh you know storage layer and kind of everything you need to be able to run your uh applications and that that uh hypervisor we use is kvn based and so we have built our own storage layer that's explicitly designed to to be consumed by that kvm based hypervisor uh which allows us to run on really really small form factors and so kind of those two things combine the introduction of you know fleet manager alongside just the the efficiency of our our software stack on hyper core makes for a really solid edge computing solution so hyper core is your the edge computing product itself fleet manager is your management tool um talk about the fleet manager and why we need that night when i think of edge computing you know the future of what age computing could be right is putting more data and more workloads and more places if unmanaged i can see that leading to chaos because this distributed computing world we have i mean if you think companies have problems managing data today when it's in two clouds in one private cloud wait till they have it in 500 edge locations right i mean this environment can get out of hand really fast so talk about some of the you know what fleet manager is and some of the the problems that it solves well let's talk about the problems with edge computing i would say they're having all sorts of problems once you've decided you can't run in the cloud and you have to run on premises that's a big decision to make uh you're starting to deal with the fact that you've got limited or no i.t expertise on site um you'll go back to that retail example you cannot ask your the manager of the grocery store to go and deal with it infrastructure on purposes it's not going to happen and so you know having the ability to remotely manage and monitor becomes essential for those types of workloads and those types of deployments and that that's really where fleet manager comes in but it's combined with a lot of intelligence that's built to the hyper core uh and so it's um you know you talked about sc platform scale computing platform as an overall solution combining those two things because it's really hard to separate them out a lot of the benefits that you have from hypercore is the fact that there's a lot of intelligence built into it that does things like monitors thousands of conditions to be able to say all right so this is my current state this is my desired state how am i going to navigate this path and this clustered environment to get to this desired state and do so without necessarily being able to connect to the internet so there's a lot of intelligence that's that's built into the solution itself that can just sit on-prem and run but if you're going to manage as you gave example 500 individual sites you need you need another layer above that to kind of manage the fleet of of hypercore based clusters out there and that's where fleet management comes in so can i think of it as a kind of a software overlay that lets you manage essentially what you would instead of a bunch of individual uh edge computing nodes you're actually delivering kind of an edge fabric right so it looks like one that's right is that a good way to think about it that's a great way to think about it it's a cloud-based offering that kind of offers centralized management for all of your deployments one to fifty thousand clusters doesn't matter it involves a lot of proactive alerting kind of gives you a centralized view of everything um you know allows you to drill down to that level if you needed to get to an individual level that happened to be experiencing an issue but it also gives you that that fleet wide view of everything that's created well so in layman's terms so for people that are sort of technical you can do for edge sprawl what vcenter did for vm sprawl does that agree with anything i think it's great yeah so so now this is a it's a relatively new product but you do have some customer deployments uh can you talk about a couple of customers and what they did with fleet manager and the benefits they saw well um yeah so we we have been working on this for a while now and so we've got a number of customers that are using it we're officially announcing it this week making it available to anybody but um but yeah with probably one one of the best examples that i can give you is with a company called ahold delays so this is a multi-billion dollar grocery retailer based out of belgium and they've got thousands of sites worldwide but originally when they came to us they were they were looking to deploy um you know their first 800 in belgium to replace the infrastructure they have out their sites and so they at the time were looking at we only had hypercore and so it was you know they needed something to be able to run on-premises buy that autonomy um to run those applications and that worked great but ultimately as they started to roll those out and got into the hundreds they needed something to act as an overlay and so we we developed a platform with fleet managers specifically for that type of use case and gave them early access to that and so far the feedback has been phenomenal i mean they basically said that it's allowed them to effectively manage these clusters across hundreds of stores to ensure that those applications remain highly available and it's saving them a ton of time and money yeah and i think um uh the best way for the people watching this to understand what the product does is you're gonna give us a quick demo right now right perfect yeah that i can talk about all you want but until you see it it's not real yeah let me give you a demonstration of the product um first things first though just kind of set the table as to what it is you're going to look at here is as i said sc platform scale computing platform is made up of really two different pieces one is fleet manager which is the cloud-based offering as you call it the the overlay to be able to manage these multiple clusters out there and then we have the software that is running on these clusters um so i guess to just visually let you see what this looks like this i took a picture with my iphone of my three node he 150 clustering this is a really small form factor cluster we've got it doesn't matter if it's this small form factor if it's you know 1u or 2u node with gpu or without doesn't matter really what the hardware attributes are our software running on these things is the same and in 99 cases when somebody orders one of these it's going to ship directly from our integrator to the end user at whatever location they want to uh so that they can they can go through the deployment i figure um just you can kind of see what it is you're looking at here start here and then i'll go into um the fleet manager user interface as a starting point so customers point to philip.scalecomputer.com they can use their local credentials or i often will just use google mainly because i cannot type on command and here i'm logged in and i i actually have this connected to our internal testers organization which has you know 19 individual systems most of which are in some pretty bad shape this is not actually how it would look right the first time a customer is going to log in it's not going to have any information in there but once they've deployed most of the time this is going to be a very green screen and everything's happy we've got some fun colors in here to just kind of show off what it might look like but you've got everything from a timeline of events that have happened kind of current health status of what's going on in the cluster even you know version info across these individual sites but it is at the cluster level when you first log in but if you drill down into the clusters themselves i've got this focused on what i call my he 150 promo cluster which is that cluster that i just showed you the picture and i'm pointing over here because on the other side of this wall is where that cluster sits um that's currently healthy um you know the very first time somebody logs in and is setting this up they're actually going to have their nodes show up in here even ahead of that on provision and then whenever they go to provision the cluster everything kind of connects into fleet manager so they can start administering things from here i'm going to click through into the cluster details and actually pull up the hypercore user interface so this is and we said there are two different interfaces here this is actually being served off the cluster itself so this is the hypercore user interface and you can see those three nodes that you saw pictured there are represented in these boxes at the top uh if this again this was the very first time you're logging in these would all be blank there would be nothing on here but you know for demonstration purposes i've got a handful of virtual machines running on here um and usually when i'm talking through kind of the pillars of of platform there are a few things i like to hit on the first is just simplicity i mean this is you can get from you know unboxing to spinning up highly available workloads on the system in less than an hour typically even less than that if you have any experience with it and i'll walk through kind of what that looks like manually then we'll talk through you know an edge deployment you're not just going to be doing this once you're going to be doing this likely hundreds of times and need you know programmatic interfaces to be able to do that just apis uh to be able to automate a lot of that uh so let me create a new vm i'll give it a name description you can select the operating system we do include performance drivers from here you're just kind of carving up whatever resources you need for this particular workload and then i'm selecting the i set which i've uploaded a handful onto here ahead of time i mean that that is yeah the very first time you put these in you're going to plug in the networking you're going to plug in power you're going to initialize the cluster by putting the ip address on each of the nodes and then you can go and create that vm just like i just did and this is probably a good time to talk about really the benefits of hyper core is that yes it is simple but a lot of simplicity uh is it's it's kind of a mask there is a there's a lot of complexity going on under the sea as you can imagine this is a cluster environment so we've got all these layers that we're managing and we have what's called state machines that are doing things like monitoring the state of thousands of individual conditions and every node in the system one of the things obviously it's going to be monitoring is the desired state of your workloads and so when i just turn this particular craig workload on that's a desired state change which means it now has to decide what to do with that and in this clustering environment where does it need to place it so it's found node 2 in the system is the best place to place that virtual machine um i can actually get directly to the user interface and see the the windows loading files here uh screen which probably looks familiar um once you once you've done that i mean that's that's it this thing is going to sit here if you've architected it correctly to run in a highly available manner out at the edge your network and fleet manager comes in to be able to manage across multiple sites in this manner uh let me actually power this off i think one of the more powerful things about this system is just how those state machines handle failures um you know that i think that's you know look at edge computing and these distributed environments and you're looking at you know what can happen in that environment when you don't have the it expertise on site and one of the big things that could happen is failures and hardware fails and that's kind of a fact of life for the fact of hardware and so you need a system that's able to tolerate those what you're seeing jump up on the screen here is this insufficient memory it's because one of those state machines that was monitoring is saying hey you started a vm that now does not allow you to fail over your vms or do things like live migrate vms around for a rolling upgrade i went and shut it back off so those have cleared themselves but in fact if it's okay with you i'm going to walk into the room grab one of those nodes and bring it back here so you can kind of see what that experience looks like for a failed node scenario and one of the customers that um that we often talk about uh it happens to be another grocery retailer which i don't need to necessarily fix on that use case specifically but this was coming into crunch time for um you know one of their busiest times of the year right before thanksgiving and they had had a couple of power issues you imagine you know retail environment you don't necessarily have consistent power consistent cooling that sort of thing in these environments and so there happened to be something in the environment that was causing node failures at a rate higher than what we would expect elsewhere and so we're on a call with them and they're kind of talking through the issue and they said we have to get this the cio talking we have to get this fixed before busy season and the director of i.t was sitting there and said what actually nobody describes as noticed and what she meant was nobody had noticed not because there weren't failures nobody had noticed because the system's architect in such a way that i can do this i can pull this is you know he 150 small form factor those who recognize the hardware it's a knock uh based platform um it's you know if something like that were to fail like i have pulled note 3 on the system the first thing that these state machines are going to do is verify this isn't some kind of networking blip this isn't just kind of transient error that that it can't recover from this actually is a situation where we cannot access this node entirely so first thing is data redundancy is degraded the next thing that's going to happen is the system is going to decide oh well that means that this node no longer is able to be accessed for serving those workloads which means fail that node restart the vms without the administrator having to you know run there and turn anything on it just handles all that for you and so the power of that means that you can put these on a ship out in the middle of sea knowing that yes the you know the cargo management system is going to stay online even if there is a node failure and you know maybe multiple drive failures over some period of time it's not a big deal they can come to port and when the port you know when the boat is docked and they have i.t resources they can bring on a new node and come and plug that in and kind of restart or heal the system physically whenever it's necessary like that so here node is is deemed offline you'll start to see those vms restart on the other nodes in the system and and that's the high availability piece of this now of course especially the edge environment this is you know when you asked what we had done specifically for edge computing um you you couldn't if you had hundreds of sites you couldn't have this user interface up at every single site right and it's just it's not gonna work um and so strong possibility you're not staring at the user interface so obviously we're going to send syslog alerts we're going to send email alerts letting people know but also when combined with fleet manager those alerts are going to populate directly in fleet manager to let the administrator know like hey there is something going on this site um and the last thing i'll touch on is just you know how comprehensive this is it's it's not just about this one site right it's yeah there are critical alerts going on the site uh but if you if you look at this we are we have a fleet of clusters out there we have a fleet of hyper core systems that that are you know maybe in various states of of need a lot of them are healthy some of them have some issues and you know the first thing somebody's going to do when they come in the morning is rest assured first of all the infrastructure's taking care of itself out the edge of the network but if they need to get into any one of these they can see kind of in priority order what is going on and kind of dig into any one of them and start you know figuring out what needs to be done so that is fleet manager that's hypercore and the overall solution again sc platform yeah thanks for the demo i think that was a a good way to uh you know to highlight kind of the power of that uh you know and uh last question you know for you craig what you know what can customers expect next can you give you know a little bit of idea into the road map of where this goes yeah yeah so i mean product management that my job is to kind of direct the road map where we're taking things uh which is twofold number one i have power to be able to do that the other thing is i can't make promises we can't keep uh i will tell you that right now we are um so often when these are being deployed in a large environment like this there's an integrator involved right and so when i was talking through here the manual steps you'd have to go through to set up a cluster uh where you're setting up the ip address on every node and yeah it takes less than an hour and you do a few of them you get pretty good on it but any steps we can take out on the manual side of things we're trying to and so we we have introduced um along with this what we call zero touch provisioning with local usb so you can kind of take a usb and put it in the back of one of these nodes that has the configuration information that you would need the next step beyond that is to be able to allow that configuration information to be available directly within fleet manager so very very soon you'll start to see some of that populated into the product where you can you can go through and set here the ip addresses that i want this specific hardware to have when it comes online so you can even avoid a lot of the you know manual steps that might have to happen going through an integrator just ship directly to wherever the final destination is it knows when it comes online here's the provisioning information that needs zero touch it's going to come online provision itself and get to a statement you can start creating those those highly available workloads good so it sounds like it uh it's even easier from here so that's right all right well craig so uh thanks for that update on you know your definition of edge i think it's very closely aligned with what mine is it was great to see see the scale platform the fleet manager product and for those of you watching the set you know i know there's a lot of uh stuff to do in it today uh the environment's never been more complicated but edge is coming there's not a you know a company i talked to that's not thinking of edge in some shape or form may not be called edge but it's very edgy as you know as you said and so uh you know just as a point of caution if you don't get a handle on kind of the complexity of it up front i think you can have an environment that really becomes unruly pretty fast and so again craig thanks for the the pre the the view of what fleet manager is i think that can really help some of the complexity that we've seen in other areas of i.t is we've rolled out computing so uh unless you have anything else to add on behalf of uh craig terriac from scale peter knows es caravalla thanks for watching zcast and don't forget to subscribe you

This transcript was generated automatically from the video's captions and may contain errors.