Ross Mauri, IBM | Open Source Summit Europe 2022

This is texturing TV. Hey guys, we're back here at the open. Source Summit Europe in Dublin.

We're talking with Ross Maury who's the general manager for ibmz, we'll be talking about Linux mainframes and sustainability. Hey Ross, welcome the show. Thank you very much.

Glad to be here. Glad to be in Dublin. It is could be worse places to be right.

I like it here. It's a complicated topic these days but to what degree RIT organization starting to base sustainability issues as part of their decisions for Designing what platforms to use her for that matter? We platform things.

What do you guys seeing what I'm seeing especially over the last 18 months to two years. So this is really recent is as corporate sustainability goals started to be put in place. Sometimes by the board sometimes by management sometimes by both that trickles down and one of the areas that needs focus and is getting focus is the it organization.

So while we've designed servers for decades to be very electrically efficient, it never really was a selling point to be quite honest now, it's one of the selling points because a customer can make real material difference in their electrical usage and really relatively short amount of time thus progressing their company towards their sustainability goals Utilization rates on traditional servers are not so high. So is that part of the problem? Is it just that we're been a little bit sloppy and how we think about workload deployments.

I think yeah. I don't want to I don't want to bash anybody too hard, but I think that utilization rates are not that high in a typical server, maybe 20 to 30 percent if you push it it'd be 40, but I think it's also the design of the servers they were designed to be commodity. They're designed to be put together quickly assembled with from different manufacturers parts and put together and a holistic design of the system really wasn't in place.

Our servers are different right we designed them from the Silicon up so we know every circuit that's in there every line of microcode and one of the things will so we optimize for performance. We're always innovating and driving higher levels of security built into this server, but we always looked at electrical usage power usage right and and we control it because again, We designed the server and end and I think that that's now so I don't want to point fingers or anything. It's just that if you're going to really take sustainability and electrical efficiency, you know to Heart you've got to start to think about the end to end solution not just a component or a card or a board.

You've got to look at the entire solution. How much progress are you making with the platform that runs Linux only versus the Mainframe environment where I can run Linux is a subset of zos and you know, which one the customer seem to prefer and why? Well, you know the zos platform is really the heart of our business and it's because of its transaction capability.

It's data handling capability and then all the other things you can do around having the transaction and the operational data there and we've seen tremendous take up of our last two generations of servers. We've seen great growth in their usage and the capacity installed globally and I think that this generation that's that was the Linux system that was announced yesterday and the and the Z version of it which was an ounce about 90 days ago. I think we're going to see great uptick with them as well continued why workloads are growing as the pandemic hit and everybody kind of huddled at home.

At least. That's what I did and we did more and more online, right you hear about the digital transformation. Well that the digital transformation that was going to take a decade I think happened in about two two years during the pandemic.

So what does that do that drives up transaction rates of everything? Of delivery of goods of paying for goods of searching for goods drives all those transactions up. So again our systems are built to sustain huge spikes in in transaction rates and things like that are systems are designed to add capacity from hands off.

They can just bring on additional capacity and those systems were exercise quite I would say quite a quite a lot during the pandemic as transaction rates grew through the roof basically, so come out of the pandemic. I still think we're going to see big fluctuations and I still think we're gonna see growth but what we're seeing now even more so than before is clients consolidating Linux workloads that may be on maybe on a private cloud or just in a server Farm or even on a public Cloud. They're consolidating a lot back on to Linux on Z or Linux one and the workload They're bringing back.

Back are two flavors one is they want to be closer to the operational data that's produced by Z whether it's an analytics workload or they're just taking advantage of again the efficiency speed and security of Z running thousands and thousands of Linux workloads all at the same time at a much higher performance rate than an industry standard server. But also as you said at the servers being pushed much harder probably run again 90 to 95 percent utilization. So it's just economically makes sense for them to consolidate from tens of thousands of servers and sprawl onto something that's much more electrically efficient space efficient carbon footprint efficient Etc.

Do you think more people are willing to have that conversation in an economic downturn because they are feeling more pressure about the total cost of it. I think the government and large companies whether it's an economic downturn or upturn have to they have to do You have to maintain some amount of momentum with their systems. They have to replace them on some basis could be could be every couple years could be every decade but there's some reason they're going to have to replace them.

And again with the huge digital transformation changing the nature of transactions. We've seen clients having to upgrade and put in new systems much more and again, bring more work if you would back to the Mainframe. Seems to counterintuitive maybe every people thought everyone was going to the cloud.

And as I tell people the cloud is in the destination. It's an architecture. And so lots of times that cloud can be easily implemented and much more efficiently implemented on a cluster of mainframes.

One of the things we don't talk enough about probably is that there's a lot of failure rates on those industry standard servers. So in the age of digital business transformation are more people looking at the reliability of the platform as well. I they are and in government and Regulators are doing the same thing as well.

And so our Hardware, you know has seven nines of availability. So it's pretty Rock Solid but our clients still will have multiple instances of these servers that can in real time shift workloads between them in case there's something catastrophic that happens a power outage or something big like that. So I think resiliency and reliability in this stage of digital transformation people are not so tolerant anymore, right they if they want to push their they want to click their Mouse.

They want to hit their phone. They want to get a response and and I can't blame them I do too. But again government Regulators are also pushing especially the financial services sector.

To really really to new heights of availability for their systems. What is the cost per entry these days because it used to be for a Mainframe. You know, you had to be in the when we call the millionaires Club where you had a million dollars in your budget for it.

Does that come down? Is it more accessible? Where are we on that curve?

So I think it depends how you want to consume it because you can consume a Mainframe in a lot of different ways you could purchase it in which case it's a larger Capital purchase. You could probably get in the game for three hundred thousand dollars not a million, but but not not a small amount of money. But there's also there's service providers that will sell you a slice of the Mainframe right virtualization the cloud model.

And so if you just want to rent the space rent the time you can do that and you can get in a very very granular level and we even have mainframes in the cloud. Now that allow clients to come in or developers to come in and like our Linux Community cloud is free a developer can get on and kick the tires and and have it so I think we've kind of lowered the barrier to entry another thing you saw. Home lending pal was here speaking on stage.

They're one of the startups in our startup accelerator. We've got more than a hundred fintex Health techs and red Techs in our startup accelerator and they're all running off of the IBM public Cloud. So it depends how you want to consume.

It depends how fine a granularity we love people in a millionaires Club, but but you don't have to be any more to get access to a Mainframe technology and a lot of people who have zos are running Linux workloads for essentially for free because the way the pricings is modeled. However, they want to use the capacity. I love it and they are you they are putting Linux on because Linux and zos complement can complement each other very well.

Especially if you're going after package applications and things like that that are just right there on Linux and you can run it next to zos because of the performance. I'm going to be being next to zos is going to be your best performance by far but also it puts it into the scope of recovery for a system. Whereas if the systems are on separate whole server instances, then the recovery of something say after a catastrophic failure and moving a workload across the country like banks are required to I think that'll be able to show that they can move their workloads in a very short amount of time and run for months from another data center to show the resilience.

So by having Linux and zos together on the Mainframe, it just makes it that much more easy to move our clothes if that's what you want to do. To your point about we have lots of transaction processing on the Mainframe people use that forever. It doesn't seem they want to rewrite those applications and yet we are accessing them from with other applications through apis.

So I guess my question is are things becoming more latency sensitive in general and that's part of the whole architecture thought. Well, that's a that's a long question you ask there. And the reason I say that is because I see clients at many different.

I would say levels of maturity. When it comes to Modern application architecture, yes, they could have written COBOL applications. They could even done it with with Punch Cards, you know, 40 years ago, right and those applications would still be running and then we did open things up by API enabling our subsystems.

So the clients could get to their applications and their data but I think in today's world what I'm seeing what I know what I'm seeing is I'm seeing some clients move to a fully modern devops Paradigm, right? They're using open source development tools whether it's visual studio and get I mean you name it. They've gone to all open source.

Technology for their devops. They've changed the way they develop applications and they're writing in golang or Java or you name your favorite language and that's how they're starting to either enhance current applications or write new ones, you know, you can have Java and pl1 and COBOL and golang all coexist very easily on a Mainframe because it do run so many different languages and I'm seeing a lot of clients again because of the productivity and flexibility of the modern devops approach and then wanting to modernize wanting to get skills right out of college that could get on and learn the tools or already know the tools. So there's a lot of Advantage.

So I'm seeing clients around the world at different stages some are still back with green screens and some are I would say fully modern and and moved into the modern era and are doing containerized workloads and running kubernetes on the Mainframe because you know, you can run containers on you can run a container platform. On Linux on on a Mainframe or you can run it within zos as well. You can run containers in zos.

So if you want to wherever you want to move your containers if that makes sense for the workload. It's that flexible. Mainframe sites and organizations a long history with waterfall processes.

Are they switching that out for devops or is it kind of live inside by side? What is the I'm seeing the the customers that are that are the ones that are really I would say Leading Edge so maybe The top two to three hundred in the world are all moving to devops all of them. I think it depends on it sometimes depends what country they're in.

What skills are available? How much risk they take and I'm not going to get into different cultures, but some cultures are are very risk adverse and others don't mind pushing the envelope a little bit, but I'm seeing I'm seeing all the leaders around the world regardless of what continent they're on moving fully to devops because again, that's the modern productive Paradigm and if you want to get again young employees coming out of college. That's probably what they're going to be used to anyway, so why would you bring them back to the old waterfall days?

So in that context is the Mainframe really just yet another note on a distributed system. And it's just one of many just a big note the big powerful secure node. Yes, that's one way to look at it.

Yeah. Speaking of security are people reviewing security of applications in general and starting to look at the underlying platform. And is that bring them back to a Mainframe conversation?

Well, it's actually interesting now that most people have realized that your data is not secure no matter how many modes and walls you put up. You know, your data is is you'd have to assume that your data is going to get breached. So what they're going to is they're going to encrypting everything and running the applications in the data in confidential Computing containers, which basically lock out human intervention even with super administrative, you know credentials.

So the baby once you lock once you lock these systems down they are just going to run and you can't be hacked and so I see encrypting all the data constantly. I see more and more more of that. I mean five years ago if there was a study done and I think it was 4% of corporate data around the world was encrypted.

I don't know where it's at least these days but the customers I talk to are trying to move to encrypting all of their sensitive data every every bit of it wherever it's wherever it's created and how it's stored. So that's one way to protect things and the next thing then is when you're running and the application. I mean, is it running in a container that you trust?

Is it running somehow containerized so that the technology will keep a hacker out and that's what that's why they're coming to mainframes because we've got that technology. We've had it for a while this new Linux one Emperor four system is actually our fifth generation of secure Enclave technology. So it's very mature and can lock down the world's applications in data.

Just about every application. We see being built today takes advantage of some form of a machine learning algorithm or maybe a yes deep learning algorithm. So where does that AI workload sit on a Mainframe?

And how is that kind of different we've been so to me there's this there's a whole Continuum from you know, analytics and advanced analytics and you move into machine learning you move into deep learning, right and we've been doing Advanced analytics on the platform for decades. But but now as clients really want to move into machine learning and deep learning you have to have more and more I think acceleration in the hardware because the software algorithms just get that much more complicated and and we'll just they will by Nature, you know, the speed of light is a limitation they will take longer to run and so you've seen specialized accelerators for AI like gpus, right which kind of you attach via a bus to a processor, but what we did in this generation and tell them chip that's in the Linux Emperor for linux1 Emperor four. Is we put inference acceleration right in the microprocessor?

So there's no going across a bus. There's a millisecond or less latency for every single inference you do and one system. Can I think can do can do 300 billion billion inferences a day.

So it's basically you've got an unlimited accelerated AI capability right at your fingertips that is a game changer. I think for clients that are going to look to Ai and the use cases we're seeing these days the natural one is like fraud detection and fraud prevention. That's a natural one.

Whether it's Payment Systems credit cards claims processing and insurance. Come come companies, etc. Etc.

That's a natural one. That's a very big payout but I think there are hundreds and hundreds of use cases that are going to get uncovered and are being uncovered as clients now think oh, I don't have to buy special Hardware. I don't have to do anything special.

It doesn't affect my performance of my system. At all to do AI I think we're going to see a much more rapid adoption of deep learning again of deep learning inference leverage on mainframes and I we've been working with a number of companies across the globe for two years. Now as we prepared bringing out these processors that are based on the telum microprocessor and they're very excited about having this capability at their fingertips.

Imagine free AI acceleration basically at your fingertips, just like we put an encryption engine in the z14 about five years ago, right? So that made pervasive encryption. You can just encrypt everything no performance.

It encrypt decrypt just pretend like it's it's a natural flow. It's not something special. That's what I think we've done to AI so I'm very excited about what's going to happen over these next couple years.

One of the biases that are out there is that people feel that the folks who have Mainframe skills are retiring. They're going to go hang out in the beach somewhere and they won't be able to replace those guys. So what is the level of skills required to manage the Mainframe environment today?

And you know, what how deep do I have to be or can I just be your average mere mortal? So it's a great. It's a great question and it's one that's been around for a long time.

This isn't this isn't a new question. Right? And I and I think we're doing two fundamental.

Things to change the game on this one is making the application development Paradigm something natural and that that everyone pretty much that's coming out of college or even out of high school is used to by using the popular open source tools for application development. So kind of like takeaway that barrier to entry because the tooling they're all there everyone be used to obviously the application logic. You have to learn doesn't matter what platform it's on but you'll take away that barrier to entry on the tooling.

On the on away from application programming then there's like system the system program that's got to set up and monitor and manage the system. We continue to do more and more leveraging machine learning and AI having the machines set up and manage themselves. Now, we're not fully there yet.

But we are we've been taking big strides to have these seat. These systems be much more self-managing than ever before again lowering that skill need for the the system program the system administrator and the final thing is what I'm trying to drive is just simplification and everything we do when it comes to the hardware in particular. Remember, the the Paradigm of the great thing about the Mainframe is that it runs applications that are 50 years old.

The bad thing about a Mainframe is some you know, some of those bells and whistles and knobs and dials that were put in 50 or 40 or 30 years ago are just archaic and so we have to go and remove them. And so what we try to do is if there's a set of tuning parameters, let's say for performance and there were they were called SRM parameters. We basically rewrote zos to have this thing called the workload manager so client could specify a policy in the English language and that would then set the performance tuning so you didn't have to be a wizard and get in there and tune it yourself.

The next step of workload manager is not setting a client setting a policy. It's the system will come with a set of With a set of models that have been trained already on the client's data their performance data and automatically tune itself in the client won't have to do anything. So that's what we're talking about.

Go from remove the old bells and whistles and knobs and make the system so that it Tunes itself. They say seeing is believing and a lot of people are still skeptical so short of confinging an entire machine. How do I go find out whether or not all this stuff works.

So there are some places that I can get us go play absolutely. Well, I mean if you want to just get on and play with Linux on Main friend, there's the Linux Community Cloud. We also have if you're if you're an independent software vendor, we have access to systems and independent software vendors can use so the developers would use some of the external clouds that we also have on the IBM public Cloud.

There's a number of different portals that you can go into and play learn. Try things crash. It's okay.

It's a test system, right? And then and then the other way is we have Hands-On Labs that we will come and work with you your site. We'll do it remotely whatever you want.

We have expert Labs that will come and work with you and we'll help you and show you the best way to kick the tires. We're happy to do that. We're proud of our technology.

We want you to kick the tires. All right, cool. Hey, you heard it here folks.

Mainframe it's not your grandfather's machine. It's a whole new world Ross. Thanks for being on the show.

Thanks very much. Thank you. All right, and we'll be back in a minute.