Drut's Composability Webinar 090823
Greetings everybody Simon McCormack here from Drut. Today's webinar is our second webinar this is about Drut composition, I'm going to do composition of a node and you'll probably wonder what that is about well throughout this session I'm going to explain it to you and show you all about it it's a word that's been used in the industry but it's actually very important particularly for the Drut technology now last time a couple of weeks ago I did a session where we did a dynamic attach reattach and detach of of resources to a previously composed unit so this is kind of the precursor to that where actually we're going to set it up first of all and it's important because there's a lot of very important technology that I'm going to show you that we have within the Drut environment that allows you to build a multi-vendor data center environment into a situation where you can basically do that Dynamic attachment and Detachment there's some dry parts of this demo where I'm going to go into building machines so you can watch that your leisure I'll try not to make it too boring hopefully it'll be interesting enough I'm going to start off with a few slides just to try and explain a little bit more about why this composition piece is actually important and interesting to you so we start off with a little slide on on droot itself and some of the messaging we've been using about environment and I think one of the key ideas and thoughts I have that seems to resonate well with customers is basically better use of your existing Data Center so what we often see is when a customer is looking at a new data center pod or Rack or or whatever they're often buying brand new servers nothing wrong with that we love you to buy new service what we have the opportunity with Drut though is to actually use existing machines as well as new machines we don't mind go either way you want I think that's the core value disaggregation with Drut is actually better reuse better use and reuse so anything new you buy you can use and make better use of it what you bought last year there's nothing worse in my mind is when you know your your environment you deployed some machines last year and then this year everyone's talking about gpus and AI imagine that you can't use the machines you deployed last year so well they're no they're no good to us that's a terrible situation to me actually with root technology we can change that we can actually repurpose those machines so we build a lot of Hardware capabilities into the solution plus software capabilities that allows you to do this uh Dynamic distribution of the environment to create it and make them more efficient and so this composition piece that I'm going to show you today is the core founding fundamental uh way that you can turn your machines into these composed nodes and then you can add the right resources that you need to and we have a lot of options with our software so we've built it in a way that you can API Drive everything so you can automate everything and we've layered the software and Hardware so there's a lot of choices in what you do so what we show you today there's a lot of options for what can use but we consider ourselves a complete system company at this point because we actually have all of the software pieces in place for you to create this brand new data Center private Cloud environment using all of our software and hardware and it can augment your existing processes it can replace some of the existing stuff it can sit side by side with the existing software as well so a lot of capabilities there now this next side I always put this one in because there's generally three key things I want to say and sometimes I forget them so it's good to have a slide sometimes I say them all in the first slide so so then I don't need to say them again but I think it's good to reiterate The Three core tenants that I put in this slide maintaining fluid resources software and Hardware that's the key thing I was just talking about make better use of what you build new or existing so I don't continue to throw old Hardware away be able to better use what you had before and make better use of what you're going to deploy tomorrow and think of it in terms of The Sovereign Hardware that you can put on it and droot software and Hardware can help you to go down this path to to making a what is a better environment and one key part of this and is the standards-based approach what I just said to you there has to be a multi-vendor I've never seen a data center that's not multi-vendor both compute storage and networking they're always multi-vendor and so she's only got one single vendor platform to do it makes it challenging right so parallels like Drut is great because we're what I'll show you today is a multi-vendor environment composing a node both new and old systems all right let's dive deeper a little bit I'm going to look at a rack when Drut we're always drawing racks of equipment because there's servers so that the Block in the middle that's a bunch of servers there I think there's networking on the top we're not going to go into networking today too much a little bit uh I've got eight servers there and the eight servers have got the drute ific in them so that they're connected to the photonic fabric so there's just eight standard type servers at the bottom there a couple of resource units with a bunch of different resources it's a big double wide gpus some little gpus some smart mix send MDB and nvme drives a whole bunch of different resources that you can build in to that that gives you a lot of advantages um when you build this kind of server so that's what you can do that little server there on the left when you compose it the demo we're going to show you can compose it add all these resources suddenly that little server becomes a big server very powerful server on demand and tomorrow you can change it make a different server add the different resources and that gives you these four key values on the left you know no stranded resources so you don't have GPU sitting machines that can't be used anymore uh we have plenty of resources so now I have a machine that maybe I only had one or two GPS before now I can make one machine have many many resources and build the impossible servers uh power options so I've got a lot more better options with power because the resources like the GPU the power hungry bid is actually not inside the server now so it gives me a lot more flexibility in terms of how I power them and savings and Power and then finally I got these interesting peripheral options so in the past I say once you've plugged a GPU into a server you're never going to open that server up and move the change the GPU to a newer one it's just never going to happen but when you put them in a resource unit actually it becomes possible I can now fit in New versions of gpus new types of GPU or new types or peripheral into the resource units and then dynamically attach them so it gives my data center a much more fluid accessibility to those resources going forward now this is all made possible with the software so there's a software slate as I'm calling it and it's not empty it's going to be filled up soon there's a software slate that makes all this possible and some of this I'm going to demo today so start off with a BMC bios and ipmi now I know whenever I do a webinar I mentioned these three terms half the audience leaves but actually it's not that bad in this case it's going to be it's important for us you as a user you don't actually have to get into it until we basically in that layer from you but it's important for composing a machine with resources obviously we need to be at the BIOS ipmi level and there's a lot of different ways to do it and you'll see in the software we have actually can be multi-bender I'm just trying to show we we need to understand at that level uh in the in the environment to be able to compose these nodes so that's step one that I'll show in the demo and it's cool and it sounds actually secondly you've got to put an operating system on your server and there's obviously in your data centers there's many ways to do this you've figured out many ways to do it uh and if it's a multivendron right you have you may have many ways to do it in your own environment so we can give you a single multi-vendor way to deploy your system OS and I'll show you that as part of the demo as well and then there's there's more physical networking on the server not going outside the server today just the network on the server when you bring up a new server you do have to do some networking configurations vlans IPS DHCP we can do all that for you with this software as well so you see I'm building up this layer of software here that is vital to your environment that we can offer as part of this solution and then the final one for today will compose our system so we'll build you a composed system that has its operating system its networking setup and we can control it at the um The BMC level uh to be able to manipulate the resources what do you do next with that system is going to come later there is more just a bare metal machine with an OS is not the end it's the beginning of a data center what comes next is a virtual platform virtual Network virtual storage virtual machines and container type environment to be able to run your workloads we've also built that into our software suite we're going to cover that in future webinars what I'm going to take what we've done today in building this composed environment and we can go into the virtualized software level and the reason we want to do that because we want to go to the next level which is the software tools I put AI workloads in there any workload it's possible the exciting one is the AI workloads but once you've composed a droop system with gpus deployed the virtual platform software the next thing you wanted to start to do is is AI workload work so that's that will show you that Journey you can go on with droot you can go on a journey with us taking you all the way to application deployment a single front-ended portal Suite with a lot of uh droot built software using the best of Open Source all integrated into a single platform comes a very exciting system level solution that we can offer you so today I'm going to do these these five things I think you know we'll we'll start with in this environment I'm starting with a an existing data center I've taken some servers I put the droop fit card into those servers and connected them to our photonic fabric so it's new or existing I think this is the the key message of the day that that we like to think about now what servers you might ask are compatible with the Vic 1000 that's a good question well most of your modern servers have been built with the ability to take a GPU and some and those gpus are big Power hungry wide modules the Fig 1000 is not as big as them so generally if you can put a GPU and a server you can put the pic 1000 in to replace it or augment it you don't have to replace it so a lot of your existing data center is already be compatible with this module and as it will show you the software we have today it's multi-dender so it'll work with that environment and I'm not against or for any vendor in that what I'm saying there is every data center I've seen is multi-vendor doesn't matter where it is even if they are a big fan of Dell big fan of Super Micro big fan of hpe you've still got some service you bought last year from a different vendor so they all become multi-vendor which is a I find that being one of the biggest challenges when I talk to Data Center operators is dealing with the multivender environment they have and the tools to use it so it's critical for us to be able to work in the environment the second piece servers discovering the servers commissioning a BMC and RPM this one I'm going to demo very shortly order automatic or manual New or Old or and old manual new and old that's kind of what it is so we're going to show you an automatic method manual method new and old bring them bringing servers into the environment so that we can then eventually compose the service before I do that I'll get into the provisioning the network and so I'll talk about show DHCP VLAN IP address configurations that you can do and you might say well don't you need to do that after you deployed the OS actually we can prep it with that software so you can prep the networking options for the machine before you've deployed the OS convenient you can change it after you've deployed the OS nothing's stopping you that but the software suite we have allows you to basically build that that provision networking piece as well and then you know that next step will deploy the operating system onto one of the systems and it can be operating system of your choice or anything that you can use and you know in any in any production environment you're going to have a mix of operating systems depends on the workloads that you have and then finally we'll do a composition of a server now what I've talked about here there's a lot of stuff to do this live I mean do it live but I when I sat back and think about how can I do this live because this involves discovering machines booting machines deploying operating systems we don't want to be here for an hour do we so I've got a number of machines in different stages that I'll show and hopefully it'll make sense I think I'm on to demo time so just looking at the clock see how much time I talked for I think I talked for more than I was planning to talk for say lovely the recording I'll cut it down to a much shorter time frame that's the beauty of recordings we can do that all right I've jumped onto this machine actually it's not the machine I wanted to start I'm going to start on this machine so this is our um what do we call it this is our DSP droot DSP software piece and I've got a bunch of machines in there we'll see this guy this is two actually I didn't rename him last time I discovered him this is three I'm going to delete them both I've done this a few times and I want to delete them because I want to ReDiscover them deleting them from the system is that it's essentially they'll be fresh to us so I'll start off with um one of them let's go add Hardware so this is the manual adding of a hardware machine I gotta give it a name this is my dsp3 now I don't need to touch most of the options I can just name it give it a power type example ipmi where you see this multi-vendor there's a lot of different options there so it's an ipmi system I need to give its IP address you can see I did this earlier right this is my browser filling this in for me and it's not right and then I gave it the password for the ipmi access that's all I need to give it and it's going to go off and basically discover and commission that machine aside machine you see it's now commissioning and powering on we can view what it's doing I mean it's got to build a machine to go to the machine it's going to take a little while so I'll start it off I'll put it on the side and then we'll we'll observe this is 130 right this is this is the 130 console access and I'll launch the console and I'll just leave it on the side and we can observe it there's some interesting times which I'll I'll jump into don't just have to log in of course and that'll that'll let me see that and then we can as we observe what's going on I can point to it and it should give us some interesting some interesting points as we're going through it and launching just the console access to this machine so we can see as it's booting and we should see it's like uh it also gives me a bit of confidence in a live demo that I know it's doing something nothing worse when you're on a live demo and doing things and you realize you forgot to press a button say there it is off going doing its commissioning on the side and obviously our software gave us that value as well now I have another machine as well I want to Auto discover I want to show that difference because the software we have here can as the machines boot we can discover them as they boot so the machine that didn't tell you anything about so my 127 it's this guy actually I'm going to go to him and I'm gonna power cycle him actually I'm gonna lose I want to pixie boot him because it'll detect him through a pixie boot because that's a login right makes sense if I want to pixie boot him so when he picks he boots it'll be picked up by the system and then automatically brought into the environment chain you can just you've got tons of machines that you need to deploy machines are actually off but I'll set up I'm gonna make him pixie boot one time that pixie booting he'll be then detected and we'll be able to bring him into the system as well power him on and I'm just going to ignore him for now and I'll carry on with the damage I've got two machines churning away doing stuff you see we detected the one that we manually added this to take a while so this is going up and basically discovering the machine it's actually booting right now once it's booted it'll go through the ipmi commissioning discovering the resources that that system have putting in the inventory and then setting it up available for us to do the next step and these machines you see on the on the right here have already been deployed in The Next Step they went through that commissioning process and we discovered all their resources I think that I can click on one now and I can show you this the environment that we have so it's learned about the physical bare metal information inside This Server knows it's a Dell server the cause memory storage networking learned all that so you've got a really powerful Inventory management tool for your data center just with this software which is actually something that's critical to a lot of environments anyway and you maintain this and this works for both bare metal and version machine so it's an interesting side with virtual machines we'll get into in a future setting now if you remember the steps I do the next thing I was talking about was so this machine's ready to deploy I can deploy the OS on it but I want to talk about a couple of things before I before I go through the deployment first of all it's a networking setup so it's discovered uh this machine it discovered all the networking the physical net interfaces that are plugged into it now if we actually look on this machine on the left we'll see what it's doing now it's it's doing its pixie boot now so this machine that I manually added it's gone through booted up and then doing a pixie boot that the our DSP software is supplying the Pixi boot software to it so we've interrupted the the boot cycle and we're giving it it's um base operating system for us to do the commission so it's progressed through that so the software we have here actually provides a lot of capabilities one is you've just learned we can pixie boot machines um back to my networking setup so I've learned the next and you see previously on this machine I configured some vlans onto it so I basically said uh no dsvp but here's a VLAN uh an IP address and subnet that you're provisioned on and you can do this to all your machines this is essentially before I've deployed the OS think about that the OS has not been deployed but I can do it before that so that when the OS is deployed it's now configured with the networking setup and they should be able to talk to the network and your application should be able to much easier to deploy them so it really simplifies that whole networking piece when we set that up now within the software itself this piece if I go into the subnet configuration and I look up my 53 VLAN for instance I can go into it and I can configure DHCP so not only is this software a pixie boot server it's also a dhp server and these are optional you can do the dhp through different resources but this software that we have as part of this true package becomes a complete system software that can use through a data center it's actually really powerful software it's built along a lot of Open Source so the DNS Master HTTP config utilities that are often used in millions of data centers around the world are part of this raw thing you can manipulate manually as well in this setup but here's a nice UI that lets you manipulate it see I've got a reserved range of addresses that I can apply when machines attached to the dfcp you'll see them listed you can statically reserve addresses for machines so a comprehensive DHCP serving environment and you not only use this for your bare metal servers because the servers themselves need addresses but also the virtual machine layer will eventually deploy you can use this to manipulate that as well making the entire system working nicely within this network environment government provided by the the Drew software that was networking let's go back to my machine because there's more you see a bunch of different tabs along here another one is storage so obviously the bare metal machine comes with a bunch of discs and you've got to put file systems on the desk so so when they boot they can deploy the software and you can manipulate that file system piece from this page as well you see all my petitions that are available add new file systems modify the disks modify the boot disk and a lot of the The Better World battle creativity that you need to do on a new system again giving you really comprehensive bare metal usage outside of the droop core value prop for disaggregated PCI devices is in addition to that you get all these capabilities with the solution becoming very powerful our PCI devices uh this is what we're going to go into more very shortly because I'm going to actually compose one of these systems into environment it gives you the basic PCI structure and so on and so forth through through the different configurations so we'll notice our um the machine I added My Demo three so that was the one I went in and I said it was a machine that was sitting there it essentially had nothing on it imagine you literally just brought it out of the uh box brand new machine racked it and all you've done is given it an ipmi address so we do need an ipmi address so using whatever you have in your system for ipmi it's got an IP address you gave that IP address and credentials to fruit and that's what we did and we said you know we then took control of that machine brought it up into this state where we can we can now basically deploy an operating system on it and we can actually that actually this is good time I can go to the network portion here you'll see it's it's all blank because it's just some physical networks it doesn't have the vlans now I can start adding the VLAN to this machine but actually I want to do the next step I want to talk about the next step is that I've got a fully commissioned machine that I can now deploy the operating system on it and it does make sense to do this staggered kind of approach uh it happens in all your data centers through your servers this data kind of approach there's usually a lap guys done some low level uh thinking to to figure out how to do it in your environment here you can do it nicely in this one's one system so I go into the deployment now for this machine that I discovered earlier so you notice here in the deployment now I get a set of uh let's go to the bottom a set of operating system options you can load your favorite operating system to the environment I've landed centops on Ubuntu onto the setup but you may have a Windows server or or some other server that you use that you like in your environment choose to release kernel level and then some more options I can actually do some Cloud init boot options so when it boots up run some more procedures essentially you you know you've got a a build that you want that works for your applications you can make this system then deploy it so we can just I won't I'll just hit start deployment on this one so he's going to go off and start deploying Ubuntu onto that machine if I go back to my machines you see them in the background he started he's noticed someone started to do that didn't and you see these machines at the bottom here I have some previously deployed machines that I that I that I created the deployed machines are ready to do the composition and we do that in a moment before I do that you see this guy's deploying Ubuntu now in the meantime our we've discovered a new one see it's marked as new this is actually my it doesn't know the name so I'm going to tell it demo dsp2 actually is his name so that's the machine that I basically I deleted it at the start and then I just that booted so that was like a you think about this your operator in your data center has got 10 new servers he's racked and powered them on the IPM is the extract them and powered them on he didn't even need to give them an address at that point it's been learned new by the software we didn't tell it anything it automatically learned it so you can see there's the options uh dynamic or dynamic or manual automatical manual for discovering machines so machines once they're deployed they've got their operating system on now they're ready to to build into you know what we call the composed node so that Drew can then do the dynamic attachment Detachment of resources I'm going to jump to this guy over here slightly different setup so yeah that was my plan with this is to have different machines in different stages because I wasn't sure the timing actually the timing almost worked right maybe I talk too much in point where I actually got the point where I could have composed it but I go to this setup here and you'll see I've got a bunch of machines here that are have been discovered these bare metal machines and I go to this machine so this is a super micro sm-30 and I go node composition screen you'll see this machine is not part of the Drew environment you have to compose it first it's saying it's basically the bare metal node it hasn't been composed I'm not going to do it here I'm going to go to our node screen because I want to show you all of the nodes there's one node already completed so this is a composed node s26 but I want to go compose node and choose the Zone I give it a name it's Simon's node hopefully give it a better name in your environment and then I go into compute blocks that are available so you think about this I've racked 10 new machines I've gone through this commissioning process that I couldn't could have done through the automatic manual so it's actually pretty simple it's like you you think about that the operators have racked the systems they turn them on they walk away when you're admin for the infrastructure arrives they're already there within this system we've already automatically discovered them they're already there prepped ipmi has discovered we've discovered all of the resources ready for you to deploy the OS and then compose it all of this can be done through API so it can all be fully automated it but it really is the steps I'm showing actually become very simple so you think about that with the composer nodes when you want to compose you better say I want to compose a node because you've now got you think about this from the the application user perspective he says I want a machine with 10 gpus and this much CPU and this many disks on it and this networking configuration of smart Mick something like that from this API call from here you can see these are the available machines so these nodes are available for us to compose into a specific mock block we'll take the Super Micro and we go next on the supermicro and now you've got the option now you can just compose the mode with nothing on it because you once you've composed it you can dynamically attach some resources you don't need to compose it again in the essence composing is kind of a one-time thing do this at start a day and the attachment Detachment can be done later in this session and I'm actually composing the node and then I'm choosing some block storage resource blocks to attach in this case I've got storage so I'll grab a couple of storage blocks and attach them it says basically I've taken this uh compute block with super micro 110p server and I'm going to compose it and in the same process I'm going to add this storage to it and if I had gpus in this setup I put gpus here so you can do it in all in one go I don't need to dynamically attach it in that case because obviously I've just done it in one session and I finally say compose the system and it then begins to go through the process to come to compose it into environment with those resources attached and it's really quick because it's already done so this part is super quick so if you think about that from the operational point of your environment a lot of it can be done automatically from machines being plugged into the network and powered on figure them to connect them to the network a lot of that can be automatic through this software so they automatically get into the point where you can compose them the actual composing then is super quick the adding and deleting resources is almost instantaneous and happens within a few seconds so that is the end of the demo that was what I was going to show today the the next session that we're going to do I think we're going to get more into the the virtual machine container storage layer to show where you can go next posing machines are a little bit dry it's actually very exciting to infrastructure people but I think it gets more exciting as you go by that stack so we want to do a little bit more of that so in my um in summary today what I've shown you I have done composition of a node into the root environment but I prior to doing that I basically showed you how easily we can bring bare metal machines into this environment and actually complement a lot of the service you've already have in your data center with the droop software really adding a lot more value Beyond just the composition of the nodes I don't know if there's any questions I need to ask for the live you know no questions to answer live that's why anybody interested follow-up please feel free to contact us at droop.io my name is Simon McCormack and it was pleasure talking to you today thanks very much
2023-09-15 11:20
