India has become key to modernising mainframes

It’s 60 years since the first modern mainframe, the IBM System/360, hit the market in April, 1964. Computers have evolved substantially since then. Personal computers came. Now, there’s cloud. We hardly hear about mainframes in tech conversations. Yet, they remain the backbone for some of the world’s most critical applications – in banks, airlines, government, retail. Because almost nothing can still match its resilience, security, uptime, and processing capacity.

Something like 70-80% of the world’s transaction data still reside on mainframes. 70 to 90% of credit card transactions still happen on mainframes.

Today, given India’s strength in software and computing, a growing part of the work on mainframes for IBM is happening here. And those leading this work – Akhtar Ali, VP of infrastructure at IBM India, and Subhathra Srinivasaragahavan, VP of IBM’s India Systems Development Lab (ISDL) – say they have taken on some of the most complex work around these systems, and there’s plenty of demand for talent in this space.

For Akhtar, IBM was his first job. After doing electrical engineering at IIT-Kanpur, he went to the US for a PhD and was considering an academic career when he ran into an IBM hiring manager who convinced him to join. That was in 1990. His first office was in Poughkeepsie, the heart of mainframes. He recollects there were some 40 huge manufacturing bases for mainframes along the picturesque highway to New York City, and the massive mainframes being loaded onto the trucks lined up along the road, was “a sight to see.”

Since then, he says, mainframes have reduced in size, increased in power, and I/O capacity (how rapidly data can be written or read). Unplanned downtime on a mainframe, he says, is 3.15 milliseconds a year. The same is in minutes for the best x86 servers.

Subhathra says IBM’s mainframe, called IBM Z, processes more than 1 million transactions a second. Google processes just about 99,000 search queries a second. “For IBM, these are serious business transactions. So it has to happen with the required security. It has to happen reliably, which is why uptime is so critical for us. And that’s what excites the people who work on this platform,” she says.

Modernising mainframes
Mainframes remain successful also because of the continuous modernisation of the platform, and this is what the India centre contributes to. Initially, it was a complete endto-end proprietary stack. But as Linux started evolving, IBM supported Linux. Now, as cloud becomes crucial for various enterprise workloads, teams are innovating to provide customers and programmers ways to operate seamlessly between the mainframe and cloud. “If some data needs to be shared with a cloud application, how do you securely give that access in a fast way? How do you make sure that the end-to-end requirements of the customers are taken care of? Those are some interesting challenges that we work from a software point of view,” Subhathra says.

The India team also works on almost every aspect of the processor. The team contributed to the IBM Telum processor, launched in 2021 and which featured its first advanced on-processor chip AI accelerator for inferencing. They are contributing to the Telum II processor and Spyre Accelerator, both of which are expected to be available in 2025. “We have successfully demonstrated that a credit card processing application can actually do AI inferencing using Telum chip to de cide whether to approve or reject a credit card transaction,” Subhathra says. These have to be done in milliseconds, and doing it on the chip, instead of on software, is more efficient. Subhathra says the same can be done for clearing & settlement, loan processing, insurance claims.

GenAI to understand Cobol
However, talent for mainframe work, even in India, is a problem, since youngsters tend to prefer building applications. Colleges are tending to teach languages like Python, and less of C that is required for building high-performance systems. Akhtar says they are working with academia and the ecosystem to create standardised material for training.

They have also used IBM’s GenAI platform watsonx to build code assistance for the Z platform to aid in refactoring, understanding legacy code, and help build new features. “There’s a lot of Cobol code in mainframes. Today, you may not get many Cobol programmers. But if AI can assist you in understanding that code, in modularising it, it goes a long way,” Akhtar says.


GN Awards 2025: Vote for your favorite Gadgets