As ubiquitous and powerful as cloud has become, mainframes still handle the back-end transactions that quietly and reliably keep the world running. But there are changes afoot.
Mainframes process billions of transactions per day of all types—tracking inventories, ensuring bank balances are accurate and up to date, reconciling stock trades, and other activities that require vast amounts of data to be processed at lightning speed with zero downtime. "Some 90% of all financial transactions take place on a mainframe," said Howard Dodd, senior analyst of software engineering strategy and practices at Gartner.
But the dominance that the mainframe holds over this type of work is weakening, said Guy Sofer, product manager for enterprise analyzer and COBOL analyzer at Micro Focus. This is because cloud is now seen by most business and IT leaders as reliable, flexible, and powerful enough to run the high-intensity workloads that mainframes were designed for—and because the cost of running workloads on the mainframe is high today compared to the cloud.
"We've seen that when customers move from the mainframe … they save between 50% and 90% of their operating expenses," said Sofer.
Trending toward the cloud
According to the Spiceworks-Ziff Davis report "2021 Cloud Trends," just 40% of workloads are running in the cloud today. The authors expect that figure to increase to 50% by 2023. Cloud spending also is increasing, from 22% of overall IT budgets in 2020 to 26% in 2022.
These numbers are similar to a recent Gartner analysis of cloud spending that predicts that by 2025, 51% of IT spending for application software, infrastructure software, business process services, and system infrastructure will have shifted from traditional on-premises systems to the public cloud. This is compared to 41% of spending in 2022.
But mainframes aren't going anywhere soon. Some 63% of respondents to a 2021 Forrester survey said they were increasing their use of mainframes, a one-point increase over 2020, said senior analyst Brent Ellis. "It's not a technology that's going away," he said.
Mainframe-to-cloud cost drivers
Clouds expand and contract based on demand, and costs follow that same pattern. Cloud customers pay only for the time their workloads are running or for "reserve instances," which are prepaid, discounted cloud time.
Mainframes are different. They are designed around the peak loads they will be expected to handle. Also, mainframe cost structures are complex.
"Customers pay for the amount of processing they do, the amount of I/O [inputs and output] they do on a lease basis," said Micro Focus' Sofer. Shops can spend "millions of dollars or tens of millions of dollars annually depending on the size of the processing that is needed."
Aside from the hardware costs, which are substantial but on par with a server farm solution designed to tackle the same workloads, there is the cost of finding and keeping an ever-shrinking pool of mainframe talent, monthly software licensing costs (MLCs), and maintenance contracts.
High-end MLCs can cost millions of dollars per month. Now that IBM is the only major mainframe company still in the business, it has little incentive to change its pricing structure until cloud is a direct threat to its business.
"Most people agree that over the three-to-five-year life of a system, the cost of hardware is not the main issue," said Graham Owen, product manager for enterprise solutions at Micro Focus. "MLC is IBM's big revenue area from mainframe and the customer's biggest IBM charge."
Other factors
But cost is just one issue driving mainframes applications into the cloud. Another is flexibility.
A mainframe application is like an old, very large octopus with tendrils reaching into every corner of the business. Mainframe applications tend to be mission-critical, written in older coding languages such as COBOL, and are tightly coupled to mainframe architectures, operating systems (like AIX and z/OS), and the business processes and logic those applications support, said Gartner's Dodd.
They also are hungry octopi. Mainframes are purpose-built to devour and process loads of data fast. Dodd calls this "data gravity."
"If you think about a mainframe application that has very high volumes of transactions globally … that front-end screen that somebody is typing in, it's almost a one-to-one mapping from a screen to a database transaction," Dodd said.
Difficult transitions
This makes simply lifting and shifting mainframe applications wholesale to the cloud, as you would with a modern application that runs on x86 architecture, problematic. It's not impossible but it asks a lot of today's IT teams and developers.
Many legacy mainframe applications have been in production so long the people who built and designed them are no longer around, so no one really understands fully how they work, what data and data stores they interact with, and what their dependencies are, said Sofer.
"Mission-critical business systems, the core business systems, were written 20, 30, 40 years ago," he said. "They're now part of an entire enterprise ecosystem. When customers move to the cloud, the first thing they need to do is understand what they have."
Typically, customers start the process by collecting all the different artifacts to map the application, he said. Then they can understand what the business needs to move and modernize. Once this is done, they can begin to plan the project and the target architecture, whether it's public cloud, private cloud, or hybrid.
Tools to help
There are tools to map mainframe application dependencies and ones that allow developers to emulate mainframes on x86 systems and emulate cloud on mainframes. This allows developers the freedom to use modern development methodologies such as DevOps and agile to build and test mainframe applications on the Linux operating system.
There's also a set of application APIs that are being developed to plug modern programming languages into mainframe architecture, giving developers a normal DevOps pipeline, said Ellis. IBM also has started to develop OpenShift for the Z series of mainframes. That means that OpenShift containers running in a cloud could also be pushed to a mainframe environment.
"You could choose which platform you're running things on," Ellis said. "You can have a Kubernetes-based containerized application that is pushed to mainframe without having any significant modifications done to it at this point."
You can also recompile COBOL code to run on x86 in the cloud either on virtual machines or in containers, said Sofer. At AWS's 2021 re:Invent conference, the company announced a new mainframe modernization service to help companies move their mainframe workloads onto AWS.
What is emerging isn't so much an either/or scenario but, similar to what has been going on in corporate data centers for over a decade, a hybrid approach. This puts things such as customer-facing functionality in the cloud but the heavy lifting is still done on the mainframe, said Sofer.
The hybrid route "used to be mostly for smaller mainframe shops," Sofer said. But "we now see the biggest mainframe shops in the world assessing, reviewing this approach to take their mainframe to the cloud," he said.
Keep learning
Get up to speed on digital transformation with TechBeacon's Guide.
Download this free IDC white paper, "Enabling End-to-End Digital Transformation".
See IDC's Futurescape: Top 10 predictions for digital transformation.
How important is digital transformation to your org? Take our survey and find out how you stand next to the competition.
Thinking of making a change? TechBeacon's Careers Topic Center provides expert advice to prepare you your next career move.
