B2B Tech Talk with Ingram Micro
B2B Tech Talk with Ingram Micro

Episode · 3 months ago

What’s Trending in Enterprise Data Centers


As enterprise customers embrace a new era of computing driven by the convergence of cloud and AI, their networking infrastructure must change to handle the sheer amount of data and traffic.

What trends are driving this paradigm shift, and how can network solutions help enterprise customers keep up with it?

Shelby Skrhak speaks with Jeff Shao, Director - Ethernet Alliances, and Bill Webb, Director of Ethernet Switching - Americas, both with NVIDIA, about:

Why NVIDIA’s GPU solutions are important for enterprise data centers

- Changes and key trends in today’s data centers

- What enterprise customers should do to keep up in the cloud era

- NVIDIA’s networking solution for data centers

To join the discussion, follow us on Twitter @IngramTechSol #B2BTechTalk

Listen to this episode and more like it by subscribing to B2B Tech Talk on Spotify, Apple Podcasts, or Stitcher. Or, tune in on our website.

...you're listening to B to B Tech talkwith ingram Micro. The place to learn about new technology and technologicaladvances before they become mainstream. This podcast is sponsored by ingramMicro's. Imagine next. It's not about the destination, it's about goingsomeplace you never thought possible. Go to imagine next dot ingram micro dotcom to find out more. Let's get into it. Welcome to B to B Tech talk with ingramMicro. I'm your host Shelby skirt talk and our guests today are Jeff Schell,director of ethernet alliances and Bill Webb. Director of ethernet switchingfor the Americas both of NVIDIA. Jeff and Bill so thrilled to have you heretoday. Thank you Chevy. Great. Thanks Shelby for having us today. Well yeah,so today we are talking about trends and enterprise data centers. But firstJeff and video is well known for its GPU solutions and gaming. But in videois also a leader in kind of end to end open networking really for all layersof hardware and software. So tell us what that means then for enterprisedata centers. Sure. Chevy, I think you're right. Media is well known forits GPS, not only for gaming and computer graphics but also inartificial intelligence machine learning and high performance computing.So in fact video Gps provide excellent computing in thousands of applicationsthat we shape all industries for example a video GPS are used inrecommended systems, voice recognition, patient diagnostics, new drug discovery,manufacturing and industrial automation or to move to drive five G. The listgoes on. Social media provides not just powerful GPS but also a full stack oftechnologies and solutions centered on ai including trips systems, algorithmsand applications. So when I say full stack, it also includes an openecosystem where immediate and partner solutions work together seamlessly inboth hardware and software. So before we dive into a media data centersolutions, let's first look at what's happening in the data centerinfrastructure. So as we all know, there are three pillars computerstorage networking in the data center which constantly outpacing each otherover years, computing processes get more powerful. Each new generations ofprocessors significantly increased performance every 18 months to twoyears. For example the latest and media a 100 GPU based systems can deliverpeta flops in performance and storage devices become drastically faster inboth our eye ops and latency leveraging the new mediums such as a VM E flash.Consequently the network connecting these computing storage devices speedsup in a much accelerated pace from 10 gig to 25 100 gig and we'll have peoplestarted using 204 100 gate. So with all these technological advances, buildinga data center actually takes a lot of heavy lifting, choosing the rightarchitecture components, deliver and operate deploy and operate um in themost efficient way. So it's not only about just about technologies incompute storage, networking virtualization and security, but alsoabout fast deployment time to market operation efficiency and scalability.Speaking of scalability so that 90% of the world data is generated in the lasttwo years and much of that is stored in data centers and new applications todayare composing containers thousands millions or even more of themdynamically added run and deleted in the data centers. And for A I machinelearning The largest models today have...

...over 100 billions of parameters. Sothere are a lot of factors and considerations in building a datacenter today into ability between old and new devices and multi side hybridcloud, integration security consumption and services models and so on. So allthis means especially for enterprise customers that complete and two anddata center solutions are very critical and indispensable. A media providessuch solutions proven in the field and for the future so that enterprisecustomers can build their data centers faster and run them much moreefficiently. So and also as data centers are in continuous change andscaling with immediate solutions, enterprise customers will no longerface the challenge of complexity of scaling also efficiency and scalability.All of these are issues that you know that are facing companies and somethingthat is especially prevalent right now is digital transformation. It's reallyimpacted every business business operations back in I. T. Infrastructurereally everything. So of course this trend has accelerated in the lastdecade with so many emerging technologies but bill what specificallyis driving changes in the data center? Yeah that's a great question and I'llkind of discuss it from the bottom up approach tying into exactly, you know,kind of the transformation described what's happening. The ability to scaleout and also to kind of maximize their infrastructure is key. It's Jeff kindof described right. You know, applications need to support more users,we need to get more throughput out of our infrastructure um in the a I worldmodels are getting larger, there's more and more data. Right? So the concept ofScale Out has really taken over the data center where we can addinfrastructure and these applications and storage can can still run at thelevel of performance required as we scale it out. So how do we build outthis scale out infrastructure? It's really focused on, you know, pools ofresources of disaggregated infrastructure so we can, you know,break apart the CPU and storage, you know, the compute nodes and storage. Wecan also um you know have pools of Gpus for Ai and and high performancecomputing and such and as applications need these resources there aredynamically allocated to them and made, you know, accessible to them, kind ofon demand. Another aspect of this is how, you know, the technology aroundhow you build and deploy your applications. It's all about leveraging,you know, a cloud native mindset, this is, you know, leveraging virtualizationand container ization of applications. This is kind of required really to tomake the most out of scale out infrastructure with with yourapplication um and really it all leads down to, you know, software defineddata center the ability to to manage and orchestrate this, this entiresystem actually the entire stack bottom to top and also end to end. So you kindof look at what what in videos, you know, really focused on with with oursolutions. You know, we want to provide infrastructure that that's um cansupport all different workloads um on the same infrastructure, right. We wantto provide a full stack solution, you know, this includes these kind of umyou know components the the you know, the hardware components to to thesoftware stack that that sits on top of it. And we also want to be able tosupport this, you know, kind of a multi cloud environment. So so let's say thatyou start a project, you can, you know, initially run it on a cloud in thecloud provider is kind of um you know, a bootstrap and then when you go toproduction you can make the decision a financial decision. You know, do youcontinue to run it on public cloud? Do you need your own on prem resources andit's a decision that you can make on an...

...ongoing basis. Um So this is, you know,some of the factors that are really, you know, leading to this um you know,this transformation specifically on the network. This is something that that isreally, you know, I kind of come from the network background in the datacenter. This is really changing the game. You know there's a lot moreserver to server communication in the data center. This is really pushing theperformance requirements of the of the network in the data center. The networkhas to tie into this software defined environment. That's critical and alsosecurities is becoming a bigger concern. We'll get to that a little bit laterbut but now because you have many different users, potentially manydifferent organizations using the same data center um security now is is acritical concern that has to be considered well. So then in in videosview, Jeff what are some of the key trends in today's data centers And Iknow that's a large question but in terms of you know what is what is new,what is emerging and I guess what's essential. Right? So you know I thinkbill kept it pretty well. So I think you know at the very high level I wouldsay cloud and Ai and the convergence of these two. So we may call this aidriven clouds. So if we look back. Right? So um every area of I. T.Solutions evolutions is triggered by paradigm shift in culture shift whichcome in with Disruptive applications and computing models and the PC andinternet error in the 1980s and 90s were with applications such as personalemails, what browsing and computerized workspace. A mobile cloud era in thefirst decade of this century was ignited by mobile phones, e commercegoogle or web searching and social media platforms that are made availableto billions of mobile users. So um back then data centers where thousands ofservers were built and very soon that size increased to tens of thousands andhundreds of thousands of servers and then public clouds like you know a Whas emerged. And so with the digital transformation in the last decades,cloud has become mainstream with more and more enterprises embracing it andbuilt about on premises. And so this start with a public cloud serviceproviders. But now we see legacy telco data centers are being replaced bytelco clouds and legacy enterprise data centers are being replaced by privateAnd or hybrid cloud deployments. So we're seeing the conversions to cloud.So on the other side um today we have a vast volume data at our disposal. Muchof the data is collected by machine such as in 5G or Internet of things.And so oftentimes these data our unstructured and without us fullyunderstanding the semantics of the data or how to use them at the point ofcollection. So all these data can only be processed and analyzed autonomouslyby machines so we can keep ourselves on top and extracting sites to make soundsand accurate business decisions. So often the value of the data must beextracted from its entirety. So models are built and trends in essential datacenter and inferences made at the age so that leads to a new wave ofcomputing where software can learn and machine can reason. So today we are inthe area of Ai including machine learning and deep learning. So datacenter has become the new unit of computing as we call it. So asenterprise customers embracing this new era, so their networking infrastructuremust be changed because of many challenges. For example, applicationsare virtualized and containerized, distributed across data centers anddecoupled from the underlying...

...infrastructure. So that results in amassive increase in the east west traffic and the need for dynamicprovisioning and management of the infrastructure. So if we look at, youknow, the network fabric, So we see clusters of faster servers drive everyincreased bandwidth which requires higher hardware efficiency from theswitches. So when servers moved to 25 gig switch, interconnect were at 100gig And as service now running at 100 cakes and higher speeds, The switchinterconnect will move to 400 gig and then manual network configuration builtfor tens of notes that doesn't scale to meet the needs of cloud computing model.So Cli based tools are being replaced by api based tools as operationalefficiency is needed to scale management of service services andnetwork devices and lastly, the reliability and cost effectiveness ofthe cloud network can make or break cloud business models. So as a resultorganizations are looking for more efficient modern network architecturesthat can be managed monitored and deployed in a scalable manner. Soemerging operational models like develops and net develops are bringingcontinuous integration and continuous development to data centerinfrastructure. Also bill when we look at the enterprise customer, the shiftsthat are happening are are, you know, really astounding and it's, it'shappening quickly. So how can enterprise customers really keepabreast of uh, this kind of paradigm shift in the data center and reallyadapt themselves to the cloud era? Yeah, that's a great question. For one thing,it's critical that enterprises do this right, because smart businesses aredoing it and to be to stay competitive, you need to understand what cloud meansfor you, what this digital transformation need means for you andhow you can potentially leverage it to your advantage. Right. If you just lookat the basics of of, you know, the infrastructures that support yourbusiness and and what it means to your business model first, you know, youhave to consider obviously Capex or or how much or how you spend to get theinfrastructure. The cloud model changes things considerably. I mentionedearlier for new projects, you can almost have zero investment in inphysical infrastructure by leveraging public cloud and then on an ongoingbasis. Make a decision, do you do you continue, you know, a consumption modeland on public cloud or build your own um infrastructure, I think um even moreimportantly it is what it means to to both op x and and how you do yourbusiness on, on an ongoing basis. Right, Jeff kind of talked about this, thisdevops model. If you look at the way um look at the software world where um the,the idea that um, you know, software is developed and continually tested, itgoes through continuous integration and continuous deployment. That model nowis available for infrastructure in the network using devops concepts. And theresult is you can do more with less, right? Everyone has demands that say,hey, you know, do more, you don't get more resources to do it. And the way todo to accomplish this is through devops um concepts. I think what's even moreexciting is is, you know, the ability to accelerate what you need to get done,the ability to get stuff done faster and, and you know, to, to minimize thetime to market, um because if you want, when the leverage, you know, thesetrends in the industry to, to your business advantage, if you can getstuff done faster than your competition, um you're gonna come out ahead. So thenthe question is, well, how how do you get started on this journey? Right, andI think we recommend you take small steps at first, but you also want tomake plans for for leveraging your work...

...for future expansion. So if you have anew, you know, like a ipod that that that you're considering or or perhaps,you know, maybe a smaller hyper converged cluster running just a coupleof isolated applications. That's a great area to, to say, hey, I'm goingto adopt a, you know, a cloud native model for my applications and for mydeployment right then I'm gonna adopt this automation first mindset. Umreally you need this automation first mindset and the thought of of treatinginfrastructure as code to make this whole devoPS model work and then is anorganization. You can start to build up the expertise, the skills and you know,the actual infrastructure and everything around it to support thismodel and leverage it for future deployments for really your entireapplication and infrastructure environment also. I mean we're talkingabout, you know, scaling out the data center and of course with the advent offive G and IOT, you know, networking plays a critical role in connecting allof this together. So Jeff, what is uh in videos networking solution for datacenters. Right. She be So at the media we provide entrance software definedand horror accelerated networking solutions for data centers. So as Billjust talked about so in the data centers now working needs to providehigh performance, low latency Interconnect to unleash the maximumpower of compute and storage systems. So you also need to scare out easilyand efficiently on to facilitate data center expansion and cloud integrationand it also needs to be open and software defined to supportvirtualization security and automation. So only with software defined andhardware accelerated network infrastructure, the data center cantruly achieve cloud scare efficiency. So the media needs all theserequirements with a complete stack of networking solutions. So in the serverswe provide connects to a network adaptor cards with hardware offloads.So we call them a smart mix and then also with a blue field data processingunit or Deep US which accelerates software defined networking software,different storage and software different security. In the fabric weoffer open switches and cables and transceivers to connect these devicesor servers. So through our exploration and open networking and to end. So oursolutions make the network infrastructure totally transparent toevery workloads and delivers a predictable services and with real timevisibility and actionable insight and furthermore all these capabilities andservices can be integrated into data center infrastructures through a P. I.S. Libraries and S D K. S. That's the very definition of software defined.You know, we've gone over a lot today and uh you know, as we start to wrap upthis episode, we always ask our guests where they see technology going in thenext year. Yeah, because they've got their pulse on one thing. So Jeff letme start with you. Where do you see technology going in the next year? Sure.I wish I had a crystal ball. So if we are if I and to pick I expect thatmulti cloud software defined automation and ai driven cloud services willbecome mainstream. So multi cloud means the cloud of clouds. Right? Includingon print, private cloud, h cloud and public cloud where all of these areintegrated seamlessly as a single cloud platform where the locality of data andcomputing is totally transparent to the user and then resources are dynamicallyprovision and applications run across...

...the cloud controlled by software. Sowith quality of service with total security and automation is given in theclouds. So as millions of containers deployed petabytes of data stored, youknow, a lot of users service dynamically the networking is inconstant change in both capacity and scale. So by replacing manual networkoperations, automation is the only way to achieve and deliver services. Soconfiguration will be programmed simulated validated by software beforethe physical network is actually deployed. So everything Wallabies apidriven. So we'll see more adoption of continuous integration and continuousdevelopment. We just talked about and later in this podcast series we willtalk about a media digital train which provide a way to protest valid,validate your deployment before it happens so that it can deploy forproduction with confidence and then a I driven cloud services really just meanthat we extract insight, you know, with a I machine learning from the datacollected in the data center and deliver services in great accuracy.Obviously all these highly dependent on networking that provide particularperformance and open flexible and scalable and reliable and secure. Soyou know, for the underlying network technology solutions. I stronglybelieve next year we will see fast adoption of CPUS 204 100 gig high speedopen networks and container based virtual networking bill. How about you?Yeah and a couple parts to it. The first part is um gPS everywhere. Um Idon't say that just because we're within video, I say that because AI ischanging the world right? And organizations are gonna you know, needto figure out how to leverage it to you know grow their business and how todeploy the infrastructure to support it. So in order to do that there needs tobe the idea of a full stack convergence. This is deploying platforms thatsupport everything. This has the foundational pieces that are providethe Gpu accelerated software, This is coupling it with well known enterprisemanagement platforms like VM ware V sphere, Tan xue for containerdeployment and orchestration and doing this in certified systems where youknow this this full stack is out of the box ready to go um you know, it'scertified, it's been tested, there's full support for it. So I think we'regonna, you know because of the growth growth growth of Ai becauseorganizations need to do it quickly and do it in a fully, you know, kind ofsupported manner. We're going to see the growth of this over the next year,um specifically on the network, I kind of touched on it earlier, you know,security is becoming a bigger and bigger concern. Obviously it always hasbeen a concern and I think in the world that we're going into the growth andconcern is even more so. So there's the idea of zero trust security. This meansthat, you know, a server or even, you know, a container can't talk to anyoneelse unless it's explicitly allowed to Jeff kind of talked about our deep usor data processing units. You know, these are are central to to providingthis this kind of zero trust security. And we're going to see this really kindof, you know, huge growth of the in this over the next year. This also tiesinto the idea that if you look at the network with the DPU on the server, theactual network fabric is becoming simpler security stuff, segmentation isnow moving to the server and the actual fabric supporting it in the data centeris requiring less features and becoming more simplified. This ties into a verykind of popular open source open networking project called sonic. If youlook at Microsoft Azure, they're the main backers of sonic, you know, it's anetwork operating system that that runs...

...on network switches, including, youknow, in video spectrum switches, um, we expect to see, you know, kind of,you know, continuing huge growth in in open networking and sonic in particularover the next year. So, so that's going to be very kind of exciting to see tosee as well. Well, fantastic stuff. A lot of great insight and uh you know, agreat crystal ball into, into what's what's happening in data centers. Well,I surely do appreciate all of your insight, Jeff and Bill, thank you somuch for joining me. Thanks should be for having us. Thanks a lot. Shall beenjoyed a lot and thank you listeners for tuning in and subscribing to be tobe tech talk with ingram Micro. If you have a question or like the episode,please join the discussion on twitter with the hashtag B two B tech talk.Until next time I'm Shelby scare hawk. You've been listening to B to B Techtalk with ingram Micro. This episode was sponsored by ingram Micro's.Imagine next B two B Tech Talk is a joint production with Swedish media andAnger Micro. To not miss an episode. Subscribe today to your favoritepodcast platform.

In-Stream Audio Search


Search across all episodes within this podcast

Episodes (352)