Jason Brown - Dell EMC ScaleIO and VMware Why and How

morning everyone thanks for coming last day early morning good show last night my name is Jason Brown I work in product marketing for Dell EMC scale IO I'm going to talk to you today about scaling VMware so anyone heard of scale IO before all right everyone oh so I don't even know why I'm here so yeah so scale IO acquisition by Dell EMC back in July 2013 its software-defined storage who service office refined storage of course so I want to talk today about why you would use scale IO for VMware you know there's a lot of confusion out there a little bit and questions about when do you use scale I wouldn't use be sin right so if you're familiar with those two technologies they're both very very similar from the outside way to get to know them there's a big differences between the two so you trying to try and help you guys understand when to use each and why we why scaler from VMware is actually a good thing so here any questions we'll keep them to the end just to keep the presentation flowing disclaimer blah blah so some industry tends first and foremost so obviously if you've heard of Sasha over to find swords you know that it's growing pretty rapidly right there's a lot of talk out there today about SDS HCI and it's coming to get the X to the expense of traditional architectures these are rays that we have been buying for the past 20-30 years and this is a study from Ricky bond who said that within the next ten years or so the SDS markets going to grow at a 22 percent growth rate to over a 50 billion dollar market in two by 2026 and that's not that far away you think about is less than 10 years away and it's gonna happen in two different arenas so obviously the public cloud is growing pretty rapidly we're very familiar with Amazon and Google and the things we're doing here right with VMware at the top part that's going to take over some of the market share what else is gonna happen in that in that hyper scale market is the companies like Facebook or Twitter they'll build their own data centers with our own commodity hardware and software and and start to move away from anything that they've been doing in the traditional manner on the bottom side enterprise servers and an HDI that's our scale io that's our visa so these technologies are going to take market share as well at the expense of these architecture's as well but it's very very important to soon there are three types of SDS in the market from an architecture perspective and they're not all the same so first of all was there something we called retrofit SDS which is really just is just a packaging exercise in truth it's really a SAN so this customer a company decided to repackage their sand in a software-defined manner by meaning that yes we can run on x86 hardware with VMs but it really don't get the benefits of SDS and HCI there's not much scalability not much flexibility you're not bringing the applications into a single architecture like you would you can see here that still those that those application leave outside so really it's it's not really the most optimal piece of SDS we feel at Dell technologies the next to something called vertically integrated this is HCI basically so basically what company decides is that we want to simplify the staff for our VMware administrators for our VM administrators so we've decided to work on a system that essentially takes the compute the storage the application all the management together into a single stack vertically and make it as simple as possible to manage so you are somewhat limited with a single hypervisor or OS but you can still do work you just you can still optimize your TCO and get a lot of benefits with multiple workloads as well that's something like V San for example now the third type is what we call general-purpose or horizontal our enterprise SDS the purpose here is to actually simplify the data center by consolidating resources and sharing resources across systems so if as opposed to going up the stack and avert the model you're going across the stack here and the goal here really is flexibility so then in this case typically is managed by the storage or server admin it can be managed by the VM admin but that's not the total focus here it's really focused on the storage but the goal is flexibility and optimizing those resources across all those Despero systems to give you a comprehensive software-defined storage platform to you not only multiple workloads but also multiple os's in height so not only can I run the ESX I could run hyper-v I could run KTM I could run the Linux I could run doc alright so you've got a lot more flexibility in this use case so the goal is here again vertically integrated like VC and simplicity up the stack sounds like scale IO more horizontal goes across the data center and the goal there is more flexibility so very important to understand that and that's helps you decide which plot which product I would use in my data center Beeson scale layout there it is so so what so what is cos Melissa you heard of scale IO so essentially what it is is it is enterprise class data center grade software-defined storage we've taken the tenants of what VMware did for compute way back in 2001 and applied that to storage abstract pool in automate so we want to abstract those local resources from those x86 servers in our data center we cannot pull it all together so that an application isn't limited to having only the local resources that it would have on a single server we want to open up all those resources that we just abstracted to the entire application and then on the back end tons of automation right tons of automation and how you manage it tons of automation and how we move storage around tons of that automation in how you add remove and reallocate storage on the fly tons of automation that we actually eliminate data migrations and tech refresh so in the culmination of all this automation is essentially a software-defined storage platform that can replace your sands they're going to replace your arrays because with that horizontal architecture that I just mentioned the entire goal is to give you an enterprise class data center grade platform for tier one applications share to applications and even next generation applications like docker like me Zoe's like kubernetes like Splunk so we're providing a high performance platform with extreme resiliency with scale i/o and actually there are customers other today's the customers that have actually replaced their entire santorini plant you know environment with scale ale so let's look at it kind of in a picture a pictorial way representative way so do you think about a server farm today you know those servers on their owns are somewhat impressive maybe they're all flash Rose maybe they're spinning destiny there's nvme in there you know with scale oh we don't really care but what happens in a VMware environment right you've got ESX clusters right with ESX plus you can go to scale is 64 nodes you can't really scale beyond that so you really usually have clusters of ESX in your environment you can manage them from a single window of course but from a physical perspective in a logical perspective you've got this kind of separation so wouldn't it be great if even though we have this logical separation in VMware clusters in our data center if we can still abside that local storage out like I said whether it's in view me PCIe SSD or even spinning disk and take it out of those servers in a virtual manner we abstract it and then we pool it all together so now for example a the VMware cluster before that was really only four servers with 250 chaos for a server now that server she has access to the whole entire pool do you by an order of magnitude you've already increased the performance capabilities the performance availability for the applications for the VMS whatever is running on those servers but that was running there before and then we had applications to it and when we had those applications the environment whether they were there before or not scale IO is smart enough to only really give the applications what they need so you're not wasting any resources here there's total optimization in the backend because of this automation that happens within the scale IO system so that there's no under provisioning there's no over-provisioning we have this big pool of storage available storage plus storage performance available to our applications and then through scale IO we actually ought to now if you provide the resources they need whether it's 10 terabytes with 400 ki office or something like Oracle or whether it's something like you know up a container platform where you need 5 terabytes in a hundred thousand ions so the ability to automate is really crucial here because yes it's great to obstruct it and give you this big pool but if you're not smart in the backend to really ensure that the system is easy to manage easy to use you know you kind of lose a lot of the benefits of any kind of STS platform and then I never mind the resiliency right so I mentioned the resiliency right there our customers today reaching six nines available with a scaling platform in VMware environments so if you really need that you know enterprise class storage but you wanted a software-defined model scale it was a very valid new platform for you so how do we do it so there's really three components of scale IO that you need to know so as I said before scale IO is installed on an OS or hypervisor so there are really what you're installing are three lightweight applications of about 500 Meg's each to create your storage system it's a client-server model first piece is the client that SDC yes the scaled data client this is your block device driver this this is this piece of software is installed on any server where you have an application or VMs that you want to access the scale IO storage component basically so you've got a server running Oracle or whatever maybe you want to sell the SDC on it because that's going to be the connection point for your applications where you present a volume to that application through the through the client second piece obviously a sense a client-server model is the server piece the SDS scale a Adina server so the SDS is important because do you want to install that on any server that you want take storage resources from so any server whatever may be whether it's one that's spinning discs in it whether has a new DME and I wonder has mixed media inside of it if you want to take some of the storage from that server and put it into that big pool the SDS needs to be installed on that server or that node whatever whatever your consumption choice maybe because the SDC and the SDS are gonna communicate together over tcp/ip they've a proprietary protocol where they talked together over tcp/ip so you're actually eliminating the need for fibre channel or I scuzzy in your data center as well because of this client-server model and because of our proprietary protocols that talk over Ethernet or InfiniBand ovaries but or Ethernet last piece is called MDM the metadata manager this is our config manager this is kind of the brains behind you know some of that automation that I mentioned before the rebuilds the rebalance is the allocation so this guy's important but we don't really talk about it too much from an eye perspective because it is outside the datapath it really is just say a component that is keeping tabs on the system it keeps the layout of the entire system the mapping and it's there to essentially say okay if a failure happens or if I add more servers or more nodes to my environment the MDM kicks off it says okay I've got something happening here let's see what it is let's start doing that automation piece that I mentioned before so let's look at scale i/o in an ESX environment anyone familiar with this picture today we've got a bunch of ESX hosts they're connected through fibre channel to a traditional storage platform whatever it may be whether it's Delhi MC or not and over the years we've become to learn that it's fairly inefficient to do this especially in VMware environments because you probably don't need to it you don't need the capacity and the performance that you may mean from these arrays because they you know are usually expensive they're usually underutilized or over utilized and they can't necessarily scale the way you want them to in a traditional ESX or even non ESX environment so as I said before scale IO client-server model the SDC runs on the server where there's an application or a set of VMs that you want to use the stores from the CIO system the backend and the SES is the stores component so very very critical here to understand that the SDC that client that block device driver that the allocations talk to that you present volumes through our lungs suit through VMFS our DMR even a direct GPIO direct path I oh that's actually installed into the hypervisor kernel there are only two platforms in the world two products in the world that are installed in the hypervisor kernel scale IO and V sin so because of our relationship with in Dell technologies we have a very very tight integration with them to be able to actually incorporate our client piece of the puzzle into the kernel itself so it's not in the LA it is actually a component that is installed into that kernel so it's very important to understand that because it is a big differentiator from other than any of the other SDS platforms of the day except for Visa yes yes while not in the hypervisor kernel is is installed as its own SVM called the scale IO virtual machine excuse me and there's abhorrence demons to service and saw this VM and that's the storage component and these guys will talk to each other over your Knicks as I said it's Ethernet as tcp/ip base so you're talking over here Nix whether it's InfiniBand or Ethernet whether it's one getting 10 gig 25800 gig grid doesn't matter to us we don't really care in the backend but the key thing to understand is that they will talk together whether they are living on the same system or not and this picture is a little wonky because there's no real sense of locality with scale IO even though Ivan saw the two components on the same platform here the the storage is not utilized on the primary server you know in a localized way the system with that automation looks to actually see ok where's the best place to put my my data right it's a to copy mesh mirror which we'll talk about but this kind of just shows you when you're inside the components where they were maybe and the space that the SDS takes up is actually up to you do you want allocate a volume you want to get allocate a disk you want allocate all the disks in the server that's really up to you so you've got a lot of flexibility because once again this horizontally based architecture gives you flexibility in how much storage you want to take from those servers or nodes whatever you want to call them so you've actually have a lot of flexibility to decide do I want to just use my own gamy drives for scalar because this cluster you want to use all of the drives for my scaling up clusters you got a lot of choice there so let's look at what it kind of looks like here this is just a representative example here where you've got SDC's installed on some ESX nodes and then even SCS has installed on other ones so it's a mesh layout and what does that mean well a single SDC will talk to multiple sdss does if you think about that if you have one application sitting on one of those ESX servers how is it going to talk to the storage right it's actually gonna talk to multiple nodes to be able to provide not only leaves the capacity but as also the performance that you may need for whatever kind of i/o you're doing so a single SDC talks to multiple SBS's and if you look at a scale i/o cluster right when you build it all out you're going to have this mass of crossed communication occurring with an assistant behind the scenes to enable that automation I mentioned before to enable that big pool of storage in the abstraction that I talked about as well so it's a massively parallel approach so all these things are happening in parallel and that's why when I mentioned before that you can run multiple applications on this multiple workloads it's because of this model here because of this mesh mirrored layout where even you can have multiple applications in your system but they're only talking to their specific guest yeses in this use case yeah they're kind of there's a lot of cross-pollination but you actually have ability to segregate that as well and segment it so you can have a sense of multi-tenancy within your platform so that one SDC could only talk to a specific set of STS is and you get the kind of security you would need in an enterprise data center there's no hotspots as well and what's really critical and crucial about this is that whenever there's a rebuild you know what disk goes down or a node goes down all those STS a--'s that are part of that one SDC link look while she talked together to rebuild that data as fast as possible so we had a customer testimonial where they had a two terabyte SSD that was failed it was 80% full as she rebuilt in 90 seconds so the rebuilds occur very very very very fast because of this model here where though all those s guesses will come together say ok I'm gonna reboot that data as fast as possible because I've seen a failure and I want to get that six nines availability that I mentioned before so flexibility also very important in terms of how you deploy it so if you're familiar with HDI right you donor to the poi one way you bet compute and storage together on the same server because your goal is to maximize the efficiency of the platform is to reduce the bottlenecks and to give you a single platform for each GI with scale il yes you have some choice again more flexibility in how you deploy it because if you think about SDS and even HCI its transformational in two ways if I want IT transformation here all weekend VMworld 2017 so you're transforming the architecture for Sur if you're booting from an enterprise storage array to software-defined storage running on x86 servers Dell PowerEdge servers for example you're transforming the architecture out of the gate right pretty straightforward but with things like HCI and even STS you're also transforming your IT operating model and that's important to understand and also very scary for a lot of CEOs CFOs folks I talk to because even though they want to get on to this train of SDS and HCI they're very very concerned about what happens in the back end what happens to my people and the people are concerned as well right if your office or our storage admin living in your own little world working on your box then someone comes into you says hey we've had an appliance now and you're gonna work on that with the server guy and the VM get the up guy there's gonna be some confusions and some you know some of this so the magic of scale i/o is that you transform the architecture art of the game but the IT operating environment can transform over time because we enable you to deploy it's a few different ways because we've got these SDC and SDS two different components that are lightweight and installed on your servers if you aren't smart enough to just to understand that where you want to install them well you can actually preserve your IT operating model so you look at the far left here this traditional two layer operating model is what you would find in a traditional enterprise data center today network admins server admins storage admins all living together and not talking to each other right scale IO compression you preserve that IT operating model with the with there's our minds with our software because if you just install the SDC that data client on the application servers the VM servers but not the SDS you've actually preserved those servers as your app servers your VM servers wherever they may be you're not taking any of the storage from those servers into that big pool of storage they're just sitting there being your compute nodes essentially and then on the bottom part it's all your STS your STS your scale iodate a server not software-defined storage you saw that on to your other servers the ones that you actually want to allocate storage from that you want to taste storage from see you now you have this two layer you've got your compute nodes with the SDC on those doing their thing running their apps running your VMs where maybe then you've got the STS nodes doing their thing just providing storage essentially so you've transformed the architecture of your data center because you have running on x86 servers with Ethernet for your storage array ascend right but you've preserved your IT operating model your server guys still do their thing your storage guys do their thing in the network guys are doing their thing as well and in an ESX environment is kind of hard to see the text here but what's unique is that would you if you were operating in this model would you use ESX for your storage notes probably not because why would you waste license on your ESX instead you'd use bare metal why don't you just use KVM or Zen or you know some form of Linux Ubuntu for example for your storage nodes so even though you've got ESX hosts doing their thing up there with their BMS or apps the storage component can actually be bare metal there metal Linux whatever may be so you're able to if may you're able to make your storage more efficient because you're not wasting your money on ESX licenses for your storage nodes because that would be kind of silly if you think about it now there are other customers that come to us and say no STS is the way each see is the UA I want to transform my architecture I want to transform my IT operating model as well that's really what hyperconvergence is all about right this is middle one here this modern or one layer co-resident HCI whatever you want to call it is tons of buzzwords for it you know my marketing guys so we've we find them all or in this case you're actually using the same nodes the same ESX hosts we're not only your compute but also your storage and this is the way to maximize your total cost of ownership minimize your total cost of ownership maximize your efficiency to ensure that you've got really the total next generation platform well this is where you also transform your IT operating model so now you your storage admin your network network admin your server admin they are working together on the system they can become more generalists so our customers that want to transform not only the architecture but also the IT operating environment this middle one here is really where they would go and that's the kind of attendants of HCI now because we're scale IO we give you best of both worlds that's one on the right a mixed configuration so maybe you're an enterprise today and you like the vision you like the the thoughts of HCI and being able to transform the data center but you're not ready to go there yes so with scale IO you actually can run in a mixed environment whereas some workloads could be that true no model that two-layer and some workloads could be that next-generation modern workload as well so you've got a lot of flexibility in how you deploy a scale IO and what's unique is that scale IO really is the only platform out there today and please correct me if I'm wrong later if there's something else out there that can give you this deployment off these appointment options and we're at VMworld we're a VMware we love our VMware friends and we've got a lot of work to ensure that the things you use as a VMware admin you can use with scale IO so we work with five five six oh six five we have a very tight integration with ESX where you can actually use vSphere to deploy configure and manage scale IO you can use the various replication tools out there the realized suite and then those features that you love as a vmware everand emotion storage the emotion DRS those are all incorporated with scale IO as well and there's more to come a lot of innovation happening within our within our company scale IO deployed on vSphere with direct path IO is coming in our next patch our next service pack which is coming out this month and then towards the end of the year early next year with our next release our big major release we're actually gonna be wall support as well where you guys have 70,000 V balls in your scaling up cluster that's actually a lot if you don't know anything about be well 70,000 is a lot I think Gurley goes to beyond that as well so as I said scale aosu scale and start at 3 nodes and scale to over a thousand knows so you actually build a very very very large scale out of cluster and it doesn't need to be just for vmware if you have vmware workloads perfect but if you other workloads as well that are running on linux windows any other type of hypervisor out there any container platform scale will support that as well and they can render with any type of media so if what's getting flexibility that horizontal architecture we talked about at the beginning that's what scale IO is because we want to ensure that whatever workloads you need to run whether it's a tier one application or a tier 3 application whether it's a traditional application or a next-generation application scale will support that so we can provide you from Hirai the IAP scan provide the performance and go by the scale that you need if you wanted to and in terms of purchasing well a few different ways to consume it as well are you do it yourselfer you like to do that build model where you wash you just buy the scale software put it on your own service whether they're Dell PowerEdge whether they're HP whether they're Super Micro whatever they may be we can provide that for you if you're at the other end of the spectrum where you actually want to just have a turnkey hyper-converged raft scale platform well VX rack flex is the system for you Dell EMC will sell you the rack the hardware the Dell PowerEdge servers the scale software the networking switches the networking cables pretty much soup-to-nuts everything and you would just say be able to consume it put in the corner turn it on and go in the middle we have the ready node which essentially is Dell PowerEdge server scale i/o software bundle with some management framework behind it to manage that life cycle because automation is important as we said but then you decide on the OS you decide on the hypervisor you decide on the network so once again more flexibility for you so whoo what does compute will uses Callao for three key workloads here as you can imagine we're talking about enterprise class storage for the data center data center in grade so obviously array consolidation blocks towards consolidation is a big feature a big use case for us you've got multiple silos of storage in your data center today it's real it's being very cumbersome to manage you are having an efficiencies in how you scale well why you consolidate those won't work loads into a single platform with SDS with scale i/o huge use case for us many customers doing that so I can use cases just private cloud infrastructure as a service and even public cloud as well there are many service providers that have bought scale i/o to build a public cloud offering for SMBs for individual users so there are many many telecom companies serve I've done that so essentially you want to build your private cloud right you've got multiple business units you've got multiple customers you want them all to come together in a single platform so you have the way to maximize your costs minimize your cost on the back end right because how does the service try to make money all they charge and if their platform is too expensive the backend doesn't really work out for them and then of course containers micro services things like that what a momentum with scale IO you know we're very impressed with the numbers we have rapid growth within our environment with our industry within our company and we're seeing you know rapidly approaching an exabyte of stores employee rapidly approaching 500 plus customers the customer does you realize the actual results customers achieving up to 50 percent five-year TCO with scale i/o so they're actually saving money on a capex side from their infrastructure perspective as well as the effects side from the right to operating model I'll go over the use cases here but really important understand that so if you want a platform that has severe parallelize parallelism that mesh mirror with massive performance and scalability as well as flexibility if you want a platform that gives you six nines availability in a software-defined package if you want a system in an ESX environment you can actually share storage across your ESX clusters and eliminate some of the silos that can happen there if you want to system that can run not only virtual workloads but also physical workloads bare metal OS Windows Linux and cetera and if you want a platform that guys seeking a lot enable you to scale CPU compute and storage together or independently scale is gonna do that for you and those are really our five super powers so I mentioned our next our next release coming up by the end of the year we're gonna do juice compression which is important Dru's V Vols if you have things as well if you like to learn more apply yourself on our website so Delhi mc.com CIO if you're interested in that V extract flex consumption model W Zak on saw slash GI tract flex so that appreciate the time I'll take questions to decide if you have any but appreciate you thank you for coming to the V Brown Bag on scale for VMware

Channel: vBrownBag

Views: 1,550

Rating: 4.5 out of 5

Keywords: barcelona, tech conferences europe, vbrownbag, enterprise education, vmworld emea, vmworld, scaleIO, jason brown, #vBrownBag, scaleIO Data, #vbrownbag, tech events europe, emea, vbrownbag Enterprise IT Education, vmworld 2017, vmworld europe, VMworld, vmware, ScaleIO, dell emc, technologies, digital transformation, Dell/EMC, vmworld barcelona

Id: s_khxwtlubA

Length: 28min 20sec (1700 seconds)

Published: Thu Sep 14 2017