Whats the effect of AI and Machine Learning on IT

02/24/2018
594
Embed

What's AI's impact on today's datacenter...and in the future? In this short video, Mike Matchett and Dave Littman discuss where automation and algorithms combine for the interative & autonomic datacenter of the futrue.

 

David Littman:                    Hi, Dave Littman, Truth In IT, joined today by Mike Matchett. Mike, welcome.

Mike Matchett:                  Ah, hi, Dave. Glad to be here again.

David Littman:                    All right. So Mike is a contributing influencer, thought leader, subject matter expert in the area of data center technologies. Today, we're talking about AI, and ways it could be used to automate and streamline data center. So, Mike, what are your thoughts on that, initially? What are you see out there today?

Mike Matchett:                  Well, what I see is a lot of people talking about machine learning, and artificial intelligence in IT. They want to do that IT operations automation better. They want to improve their processes, and they think, "Gee, if I'm going to run computers, I can maybe just bring in a computer to run the computers, and step back, and it'll take care of itself," in that artificial intelligence nirvana.

                                                      The truth is probably a little bit more practical, in that machine learning is something that can be applied to speed up the recognition of processes and patterns, and take actions proactively, and really help get towards that more lights out data center kind of concept. We're not fully there yet today.

David Littman:                    Okay. Do you think there are specific types of data centers that might benefit from AI more than others?

Mike Matchett:                  There are definitely data centers that are looking at scale and speed as critical parts of being competitive for their business. Part of scale might be looking at, say, containerization and dev ops. Instead of looking and trying to manage 100 processes, or 1,000 VMs, they might now have to manage hundreds of thousands of containers. At that scale, jeez, it's really hard, as a human, just to even comprehend looking at all that data, much less reacting in a timely enough manner to make a difference.

                                                      I think, also, part of that same question are, data centers are looking at hybridizing. It's hard enough to take operational system management tools, and look what you've got in your data center, and get your hands around it. But when you start changing your operations to spread across one Cloud, and then another Cloud, and you've got a multi-Cloud situation, and data center on presence equipment, and data's flowing all the way around there. It's going to be almost impossible for a human being, to, again, find the problem in the haystack, find the needle in the haystack, and react fast enough to really make a difference in fixing it before everything goes south.

David Littman:                    How does today's IT professional, data center professional, begin? What's the best way to start with an AI automation program?

Mike Matchett:                  Well, you definitely don't want to just hire a data science team, and cut them loose, and say, "Go build something, and come back to us." I mean, we've seen lots of companies try that, and just get really disappointed. What you're going to find out today, though, is, a lot of the vendors that you know and love are starting to fold things that are called machine learning into what they already offer.

                                                      Just about every piece of infrastructure I've seen that comes out now has a Call Home Support. On the back end of that Call Home Support, that company is probably running some machine learning for you. You can just start to take lessons from that, start to be curious about how they're doing it, and see if you can fold some of that application back into what you're doing onsite.

                                                      You definitely want to get up to speed on what machine learning can do for you more directly, but the real way to take advantage of this is, try to bring some piece of it in-house, run it in parallel with what you're already doing today, and just compare the results. Don't obviously flip the switch, and crank over to one, versus the other, and stand back, and ... That's like, getting in the self-driving car, and then scratching your head, why it ran into the bus. We're not quite there yet, but we're getting there, and so, you definitely want to be looking at self-driving cars.

David Littman:                    So, along those lines, what are you seeing, in terms of unforeseen problems and pitfalls that can arise out of AI-driven data centers?

Mike Matchett:                  So, right. So, yeah, one of the things with computers is, we know, if we cut them loose on something in operations, and we say, "Go fully automatic," they can just go from bad to worse faster than you can, sometimes. There's lots of problems with IT system management that we're going to have to overcome, if we want our machine learning to be fully automatic. One's data, data quality, data validity, security validation.

                                                      There's problems with what the computer can learn. I mean, here's an example, Dave. So, a machine learning process can really learn from what it's trained on, what it's seen in the past. If you create a whole new scenario, and then, something is going wrong with that, your machine learning algorithm isn't necessarily going to be able to recognize it, because it hasn't seen it before.

                                                      Now, some people say that is emergent behavior in some of these algorithms, where they can spot things they weren't specifically trained on. But when you really peel apart and look at what they can spot are the overlaps between problems they actually were trained on, and finding, what kind of a commonality that they can recognize in a new problem, and say, "Hey, this was common with problems in the past."

                                                      They really still can't learn something they've never seen before. So it's that Black Swan problem, if you're familiar with Nicholas [Tassim Talb 00:05:07], and Fooled By Randomness, but there are increasing ways to take advantage of scale. We're going to find that management service providers, these are people who are going to remotely manage systems on the behalf of many clients, are going to be able to compound and compile information across multiple data centers, and multiple clients, and many thousands of nodes, and many hundreds of thousands of things, and grow their expert knowledge base much faster than any one company can. And they are already doing that.

                                                      So that's good news, because that means, more and more of the possible problems that could ever happen are being captured already, and being trained on.

David Littman:                    Okay, fabulous. Well, Mike, I really appreciate you taking the time, to speak with us today, and help us learn more about AI, and the future of today's data centers. So, thanks very much.

Mike Matchett:                  Ah, you're welcome, Dave.

David Littman:                    Okay. Thanks again.

Mike Matchett:                  Yup. Take care.

 

Channels:
Tags: