By all accounts, the merger has been successful, and performance has improved, said Francis Feldman,
"We're pushing product out substantially earlier than before … which gives us a greater window to support additional applications," he said. "We are more than pleasantly surprised with the results."
That doesn't mean Feldman is completely tickled about saying goodbye to big iron.
"It's been a long process and we're actually here and it's like, 'Wow,'" he said. "It's kind of sad on one hand being an old mainframe hack. But it is a project that came to fruition and we recognize that it's a positive thing."
Though migration off mainframes does happen, most mainframe shops stay on a current platform or increase investment in it. In a purchasing decisions survey that SearchDataCenter.com conducted last year, only 8% planned on decreasing their investment in the mainframe in 2007 compared with 2006. Still, plenty of shops have moved off mainframes, usually because of cost issues.NYSE's protracted migration
The live migration began a year ago and was originally scheduled to be complete by the end of last year, but business-driven change requests pushed the timeline out as SIAC worked to ensure that it went well from a technology perspective. NYSE went from a mainframe Sysplex -- Feldman wouldn't specific how many mainframes were involved -- onto IBM Power-based AIX servers and x86 Hewlett-Packard Co. servers running Red Hat Enterprise Linux. Both environments were set up across two data centers to account for backup and recovery.
In addition to performance, Feldman said NYSE underwent the migration to save money. Though he wouldn't detail how much money the move has saved, he said it was "serious financial savings."
Trades executed on the floor of NYSE were matched and sent by mainframes to the Depository Trust and Clearing Corp., which clears and settles the trades. At night mainframes handled batch processing and supported regulatory requirements. But Feldman and SIAC found that the costs of running big iron, which includes the manpower, software licensing and services, were prohibitive.
"Within an hour" on AIX and Linux, "we can transact more business at a lower rate," Feldman said. "It made sense for our transactions to port and run them in an open environment as opposed to a mainframe."
Training mainframe staff on Unix and Linux
SIAC had people on staff that were proficient in open systems -- as it already ran them -- and staff proficient on the mainframe. The goal in the migration was to get the mainframers up to speed on Unix and Linux environments.
Much of the transition involved moving from scripting on Job Control Language (JCL) for batch jobs to equivalent scripting languages on AIX and Linux, using the open systems version of the mainframe's Interactive System Productivity Facility (ISPF), which manages and monitors application jobs, and dealing with Unix and Linux command-line interfaces.
"We had a team in place that was already proficient in open systems," Feldman said. "So we had a group to lean on when we needed some handholding."
Mainframe migration advice
Feldman said SIAC benefited from two major things in its transition from mainframes. First, the proof of concept. Feldman and his team went through the process of estimating that the migration would reduce the cost of transactions by half and improve performance. They then had to present the plan to executives and prove it could be done.
"It allows one to gauge the applications that are being targeted for migration," he said. "It's extremely important that you define financial metrics showing that it is more palatable to run on open systems."
Second: Test, test and test some more. Feldman said you should build "as robust a set of test scripts as you could imagine." Any organization migrating off the mainframe could always use more time to test -- saving money is one thing, but if your performance on the new platform isn't up to par, that's equal or more money lost in wasted time.
SIAC could have continued on the mainframe, outsourced to another vendor or tried to rewrite the code. But the group believed that moving to another platform was the most feasible solution. Then came the hard part: finding out who could move its JCL and COBOL logic to a distributed platform. Feldman said there are plenty of good mainframe migration companies out there, but some have had difficulty in grasping the difference between a smaller migration of 250 million instructions per second (MIPS) and one the likes of the NYSE's.
Some mainframe migration firms, for example, are primarily emulator software companies that Feldman said would incur performance delays after a certain number of MIPS. At NYSE, though, even a split-second delay is costly.
In came Oakbrook Terrace, Ill.-based Clerity Solutions Inc., which focuses on legacy migrations. To test Clerity, Feldman and his team collected some batch and CICS application code, sent it to Clerity and said he would be at Clerity's development center in 24 hours to see the results. Clerity passed.
"Testing enables a very solid comparison and review of outputs in order to speed up the overall migration process wherever possible," Feldman said.