Korean insurer Samsung Life Insurance shut down its IBM zSeries mainframes, shifting a 7,000 MIPS workload, including loan, contract and payment systems to two HP Itanium 2-based Superdomes running HP-UX.
The insurer undertook the massive project with the help of Seoul, South Korea-based middleware provider TmaxSoft Inc. Samsung used the company's OpenFrame and TJES products to replace CICS and JCL, respectively.
According to Sangho Yoon, director of the information strategy team at Samsung, the decision to move off Big Iron was strictly financial. "The initial driver was cost. In Korea, IBM is the sole provider of maintenance services on the IBM zSeries mainframes, and hardware and software maintenance costs are not competitive in the local market," Yoon wrote in an email correspondence.
According to the press materials, Samsung expects to save $20 million over the next four years, followed by $10 million in savings annually -- compared to what it would have to pay to support its previous system.
Where will Samsung recoup those massive figures? According to Yoon, the bulk of the cost savings will come from the reduction in hardware maintenance costs and software licensing and maintenance fees. "Manpower costs can be reduced as well since we no longer need mainframe operators, admins or system programmers," Yoon wrote.
While it's no secret that mainframes are expensive to run, one of the main inhibitors to getting off the platform is the risk and headaches involved with the project. But Samsung's story is fairly painless. Yoon wrote that Samsung turned the power off on its mainframes one month after the project went live. And the entire project rolled out as planned, in less than one year.
Yoon said he did face challenges in completing the project. For example, verification of the migrated data and code took more time than the actual migration. The conversion of double-byte Korean language data fields proved more of a challenge than Yoon originally expected. And he uncovered numerous bugs in the packaged solutions and had to wait for patches or find workarounds.
Yoon also wrote that matching the mainframe's batch performance was a challenge. In addition, he was constantly dealing with the doubters and naysayers, even up to the final weeks of the project.
But the huge project was completed with minimal snags and it reopens the discussion on whether mainframe rehosting is a viable option.
Another look at mainframe rehosting
Options for moving off of the mainframe have been around for years, but hardware improvements and platform-independent middleware have made it a more attractive proposition. Plus, as more companies complete these projects, prospective data centers have more test cases to consider.
At the same time, IBM has taken a cue from its users and has made some cost adjustments to make the mainframe more competitive. The company has introduced new ways to use the machines through new processors designed to run applications on Linux or Java, which can save companies software licensing expenses.
Hardware. According to Sean Murphy, vice president of marketing at TmaxSoft, the reliability and functionality of a rehosted Unix system are essentially the same as on a mainframe.
But Gordon Haff, an analyst with Nashua, N.H.-based Illuminata Inc. doesn't necessarily agree. "Comparing mainframes to Integrity, big Unix boxes -- it's difficult to make a direct comparison," Haff said. "Certainly systems other than mainframes have made incremental advances in reliability and performance, but I don't think anything has really changed the scene in a fundamental way."
Furthermore, some workloads don't do well outside the mainframe. Joe Clabby, president of Yarmouth, Maine-based Clabby Analytics, said hardware replacement decisions, and specifically Samsung's choice, would be based on usage patterns. For data intensive applications dealing with very predictable workloads, like encryption or data mining, Clabby said the HP Itanium method of piling data up front and then heaving it at the processor works pretty well.
But with varied workloads, it's a different story. According to Clabby, IBM's Power processors and its CMOS-based mainframes run circles around the HP Superdomes in tasks with varied workloads versus a targeted, predictable floating point.
"This customer must have had the right application mix to warrant that move," Clabby said. "You've got a customer with the sort of application that is very predictable in nature and has decided to run it on the processor designed for predictable workloads. That's where the heart of your decision is."
When it comes to deciding whether to migrate, application size matters. According to Mike Kahn, managing director of Wellesley, Mass.-based Clipper Group Inc., users that have a small number of legacy apps may be better candidates for migrating to another platform.
"Large applications that consume a large number of MIPS don't partition themselves automatically to run on smaller platforms," Kahn said. "Also, the applications, because they are that large, the complexity can make them less of a candidate to automating one platform to another."
Operating costs. The other factor is software cost. Software licensing is still a pretty big chunk of cost on Big Iron. But IBM has had some success moving Linux and Java workloads onto the mainframe, allowing customers to run apps on open source platforms.
In fact, Clabby expects the mainframe to gain customers in the coming years, thanks to these new workloads. "IBM has done a lot of pricing activities on the mainframe," he said. "Mainframes are supposed to grow 1% to 2% in market share over the next few years, but I think they could grow as high as 7%."
Staffing issues. Finally, there has been some concern recently about the shrinking pool of qualified mainframe people -- from both IBM and its user base. With many mainframers retiring soon, and not a lot of people coming in behind to replace them, some companies fear that it will be difficult to find hardcore COBOL programmers.
Recent email from readers, however, suggest the opposite is true. "If there is such a shortage of mainframe experience in the marketplace, then how come so many of us are out of work and busy cross-training in this Unix crap?" wrote one unnamed reader in response to a recent article.
"Skills exist in response to demand," said Illuminata's Haff. "You may have some local fluctuations. If everyone wanted to write homegrown custom apps and they wanted to do it in COBOL, and there were lots of jobs for COBOL looking into the future, then lots of COBOL programmers would exist. But that's not the case."
Let us know what you think about the story; email: Matt Stansberry, Site Editor