Understanding Infrastructure Automation

Machine-generated from the episode audio and not hand-corrected, so names and technical terms may be imperfect. The audio is authoritative.

Hey, how's it going? You're listening to yet another episode of the CTO advisor podcast, reflecting on some thoughts from our virtual event in 2020. Today I have Nick Brackney, who via Dell Tech, he's a director at Dell Tech, Dell Tech Lesher sponsored some of the CTO advisor virtual event, but this isn't a sponsored podcast. We're going to go right at Nick and ask him some pretty tough questions. Nick, first off, welcome to the podcast. Great to be here, Keith. It's always great talking with you.

Nick, we know, and this is something that you've talked about for a couple of years, ever since I've known you, the key to successful enterprise IT today, we're very aligned to this, is automation, being able to automate the things that are not, I don't want to call them not differentiating, but things that we're no longer building new apps on, but they're still paying the bills. We absolutely need to do performance monitoring. We need instrumenting to do that performance monitoring. We need to provision new workloads as we need to do capacity management, et cetera, et cetera.

The list goes on of all the things we need to do in IT. You work for Dell Technologies. Michael Dell has been out in the front talking about the autonomous data center, but this is hard, man. This stuff is hard. And I guess my first question to you is why is it so hard? Isn't this known stuff? You know, it really is. And I think it's, when you think about this, and we've created a new framework to think through this because we said, look, you know, there's value at each level, right?

And think of it like an autonomous vehicle. You know, I recently purchased a new car and the levels of technology that I'm seeing in that car are dramatic and amazing. However, they're not the same as being able to take my hands off the wheel and let the thing drive down the highway and feel comfortable, right? But they make things incredibly better. And so I think the first thing we have to do is recognize, yes, it's really hard to really full-blown autonomy and building a completely, you know, hands-off data center.

But along the way, there's so many incremental gains. And, you know, the first level of autonomy, we basically define it as scripting, right? It's been around forever. People have gotten used to it. They've got Excel spreadsheets with their common scripts for setups for configurations, and they copy and paste them in. And, you know, I worked at NSI for a while, and we took those scripts, we kind of cut them up, and we assembled them as pieces and created recipes and run books.

And that allowed us to get further with the levels of automation there. So I think it's hard. It's always going to be hard. You know, our CTO says that, you know, it's probably a harder problem to solve than the autonomous vehicle because the possibilities and the permutations that you're going to have in your data center are endless. Yeah, I love the background because I remember, as I went from job to job, I might take scripts from one job to another job, or I may even get kind of a jumpstart kit from a enterprise IT vendor to say, hey, you want to automate the provisioning of storage to VMware, and there will be some script that somebody wrote, whether it was a professional services firm or somebody on my team.

But it wasn't always as simple as just taking that script and reapplying it because I found that, one, every enterprise IT infrastructure is kind of a snowflake. You know, VMware is VMware from one company to another one, but the way someone chooses to implement VMware and the underlay is all different, and it's enough to break it. So we've established what the challenge is. We know it's hard. What's being done to fix it? Because at the end of the day, we both agree automation is key to being able to adopt common technologies such as Kubernetes, Cloud Native, and whatever comes next.

Yeah, so I think when I define autonomous operations, I break it down into two simple categories. One is having access to the data, right? If you don't have that, there's not much you can do. There's not much you can do. And then the second part is the ability to act on that data. So that's the instrumentation. So on the data side of the house, I think we're seeing way more data sources and way better data sources than we've ever seen before.

You know, AIOps has really become a thing that most organizations have embraced. You're starting to see even the virtual infrastructures are being created in a way that it's easy to assemble and get telemetry data from all of your various different machines. And so we're really doing a much better job of delivering better metrics, better data sets, cleaner data sets. So that helps with that first part. Now, the second part, instrumenting, right? You know, I think people have really embraced API-driven architectures.

We're really looking at certain tasks. For example, things like cash management, that's already at level five. We've taken that in a lot of our products out. You know, people don't ever have to worry about cash. It's just being taken care of, right? And I think that's kind of how we're starting to do this is as we move through this, we're going to take pockets. We're going to see small gains. We're going to see little things that are easy, things like battery life and cash and things like that.

And we can get those solved. And some of the other ones are going to take a little bit longer, the more complex ones. So let's pick apart kind of this AIOps and data thing. When we say data on this particular episode, we're not talking about data as in the SAP data. We're talking about the metadata around your operations. So log files, telemetry data, performance information, that type of data. And that data has come in all different types of formats.

Intel has this format for giving you data off of its processor and chipsets. AMD has its way. Cisco has its way. VMware and even Dell Technologies. When we say x86 is x86, that's not always true. Yes, x86 is compatible from a binary perspective. But when we're talking about operations, just getting a model number across seven or eight different server models is difficult stuff. Like if you've ever had to do CMD stuff, it's difficult. Where have you seen the maturity in this over the past few years and being able to because as we get this data that's instrumented, how are we able to normalize it first?

Yeah, I mean, on the Dell side of the house, here with CloudIQ, we've basically focused on making sure that first we get all of our different products lit up for that. So that's kind of the North Star for CloudIQ is making sure all of our stuff can be monitored from a single pane of glass. Now, that being said, people have heterogeneous environments. And so I think there are other organizations that are out there that are building Switzerland. They're building systems that can manage and they can ingest data from these various different collection points.

Because to your point, I don't want to spend 13 different consoles to get that. So I think that is one way in which it's definitely improving is that we're starting to see more and more people that are aggregators of this data and able to visualize it in a more efficient way. And let's talk a little bit more about that heterogeneous problem and opportunity as well. You know, as I look across my infrastructure, the CTO buys a hybrid infrastructure. I have some Dell servers.

I have some HPE servers. I have some Cisco switches. I have NVIDIA switches. I have Arista switches. My infrastructure is about as heterogeneous as it can be. I don't have the skill of operations from an application perspective. But as I look at it, I scratch my head and I say to myself, OK, I want to make this autonomous at a level. Where should we start to think about entering the autonomy part of the conversation, like from a management perspective?

Yeah, I think one of the easiest ones, and you'd mentioned VMware in the past, is since you're so vendor agnostic and you have so many different vendors in place, the easy part would be the abstraction layer might be the easiest way to move forward with that, right? So if I have VMware across all of it, you know, there's a lot of things I can do with NSX. There's a lot of things I can do with vSAN or vSphere, and I can at least get some of the data, some of the metrics for that, right?

The other thing that you didn't mention is that, you know, those are different vendors, but what about different sources of data, right? So log files can be changed. So log files only give me one part of the data. You know, a company I used to work for, we were big on wire data, which is the network monitoring function. And I'm not talking about packets and I'm not talking about going into Wireshark, but we were actually stripping metadata straight off of the L2 through L7 communications, right?

So that would give you one set of data, but, you know, that has its own drawbacks because it can't tell you what's going on locally on the machine. For that, you need, you know, the telemetry data and stuff like that. So I think that's one of those other opportunities is thinking through what are my data sources? How can I aggregate them and how can I use them for specific purposes to reduce my, you know, triage and troubleshooting? Yeah, so I love this.

And again, it doesn't necessarily have to be VMware, but this abstracted data center concept, as we look at the examples from the public cloud, we can't do everything that the public cloud can do, obviously, because it's a single platform at the end of the day. But we can look at the way that they attack abstraction. Abstraction where I don't care what network hardware they use underneath, but let's switch the conversation to more of a Dell-centric conversation, because when companies come to Dell Technologies, they come there practically because you guys are just so big.

You can meet the needs across the horizon of IT infrastructure needs. So let's talk about best case scenario, which is not necessarily achievable, I think. But let's talk about Greenfield. If you could have an all Dell tech stack, what are we looking at from an autonomous data center perspective? What can Dell offer customers today? Yeah, I mean, I think that probably our product that would be the highest in the form of autonomy would be VxRail, obviously. It's an HCI product.

We're working with VMware, and if you're using VMware Cloud Foundation on VxRail, we're delivering a curated experience for patching and upgrading, right? So we're doing all that testing on the front end. You're getting the payload at the end. You click download. It's cluster aware. It'll just cycle through all of your machines and really reduce the amount of time for updating. But every single one of our PowerEdge products, our PowerBranded products, there's things like iDRAC that are available for them.

We've been doing automation for a decade or two, and so we have the ability to really reduce the amount of time spent on deployment, for instance, by something like 80%. So I think it's built into everything, first off. VxRail is a foundational, really strong, autonomous operations type of product. And then CloudIQ, it's our visibility. It's our monitoring solution. It's the AIOps product, and it's free. If you buy Dell Technologies, you can get CloudIQ and light it up for all of your Dell Technologies stack that you have.

So I think that's really how I would look at that. And then, of course, storage. We have great solutions across the board there. There's a lot of automation in there. A funny story is sometimes customers don't always understand the automation that's in there, right? I remember there was a story where we had a customer that tipped over their iSalon, which is now PowerScale, and we tried to figure out how they managed to do that because it didn't make any sense.

And it turned out they had set something like 10,000 different policies because what they were doing was every single student that came in, they were giving them a free 50 gigabytes of data for whatever they wanted to use it for in perpetuity, right? So this is a college. And over time, they just keep adding in more and more people. And we're on the call with them, and we're like, where are all these policies? And they explained it to us. And we go, you realize you can just do that one time.

You can write one policy that says anyone with this email address to shoot them 50 gigabytes. That's a level of automation that, unfortunately, the customer didn't understand. But once we were off that call, that made their life a lot easier. Yeah, we actually took in VxRail into the CTL-Visor hybrid infrastructure and did some research around it. One of the things that we noted it for via the competitive solutions was the integration with VCF. It was literally a button click when you go to deploy VxRail.

It says, are you deploying? Or when you go to deploy VCF, it says, are you deploying this on a VxRail? And the level of automation is there. 0. And the consultant came back with a quote for 80 hours. And it was 80 hours of just researching patches, firmware, all the minutiae of enterprise IT. And this was my rock star consultant. This is the person that I put on to say, hey, I need to deploy Kubernetes. Should I use OpenShift or VMware Tanzu?

Or should I look towards EKS? And how do I get it to talk back to the data center? And I'm potentially spending 80 hours of his time, not necessarily just the money, but 80 hours of his time doing a simple upgrade, one major version of vSphere to the next. And to your point, VxRail did meet the criteria. Just a couple of button clicks and automation and it went away. So, Nick, let's spend the last couple of minutes looking into the future.

What can customers really reasonably expect to achieve via autonomy today? We talked, you gave the comparison about, you know, level one through five, like where are some of the best organizations at on that scale? Yeah, I mean, I think the way we set this up, and we have some documents that can talk through this, but the way that we set this up is level five is very aspirational. It's way out in the future. It's a North Star. I would say most organizations probably are between two and three.

Two is when you start to get into some more policy-driven stuff, you know, so self-service and things of that nature. Three is a little bit more behavioral, right? It's like when my system sees this behavior, it responds. And so we see pockets of that with, you know, things like I think app defense with VMware, I think is a good example, where you're starting to shift from a safeless to moving towards more behavioral. I think that's really the key is the machine starts to, within, you know, some guardrails, starts to understand and really take direction and steer.

But, you know, again, it's important to note like, you know, we saw with Fastly a while back, you know, automation, the human needs to remain in the loop because there can be an issue and automation allows you to scale an issue. And so you need somebody to make sure they're on the other end that's double checking and verifying, hey, hang on now. I'm not just going to provision 10,000 VMs right now, right? Automation does scale stupid, is what I used to say.

So, Nick, I'm going to let you go. You have to head to vacation at the time of this recording. So once this hit, people will be able to at you all they want about comments around the feedback. How do people, what's your Twitter handle? Just at Nick and then Brackney, B-R-A-C-K-N-E-Y. Keep it simple. All right. And if you want to DM me, if I went too easy on Nick, if I should have really kind of drove home that I have some HPE stuff in my data center, you can at me at CTO Advisor on Twitter.

com is the website. Rate the podcast. If you enjoyed it, and even if you didn't enjoy it, go to your mother's, your wife's, your husband's podcatcher, add it to their playlist. If anything, our soothing voices will put them to sleep. Talk to you next CTO Advisor podcast.