A survey from Compuware Corporation on attitudes of CIOs toward mainframe outsourcing showed a significant level of dissatisfaction with one or another aspect of mainframe outsourcing. Check out the survey here.
Mainframe outsourcing has been a fixture of mainframe computing since the outset. The topic is particularly interesting in light of the recent piece DancingDinosaur posted on winning the talent war a couple of weeks ago. Organizations intending to succeed are scrambling to find and retain the talent they need for all their IT systems, mainframe and otherwise. In short, they need skills in all the new areas, like cloud computing, mobile access, and most urgently, big data analytics. In addition, there is the ongoing need for Java, Linux, WebSphere, and CICS in growing System z data centers. The rise of z-based hybrid computing and expert integrated hybrid PureSystems to some extent broadens the potential talent pool while reducing the amount of skilled experts required. Still, mainframe outsourcing remains a popular option.
The new Compuware survey found that reducing costs is a major driver for outsourcing mainframe application development, maintenance, and infrastructure. Yet multiple associated costs are frustrating 71% of CIOs. These costs result from increases in MIPS consumption, as well as higher investments in testing and troubleshooting due mainly to poor application quality and performance. In fact, two-thirds (67%) of respondents reported overall dissatisfaction with the quality of new applications or services provided by their outsourcer. The source of the problem: a widening in-house skills gap and difficulties with knowledge transfer and staff churn within outsource vendors.
Compuware has published a related white paper titled, Mainframe Outsourcing: Removing the Hidden Costs, which expands on the findings from the study. The company’s recommendations to remove the costs amount to reverse engineering the problems revealed in the initial survey. These include:
- Utilize MIPS better
- Explore pricing alternatives to CPU-based pricing
- Improve the quality of new applications
- Boost knowledge transfer between outsourcers and staff
- Measure and improve code efficiency at the application level
- Take advantage of baseline measurement to objectively analyze outsourcer performance
The System z offers numerous tools to monitor and manage usage and efficiency, and vendors like Compuware, CA, BMC, and others bring even more.
The MIPS consumption problem is typical. As Compuware reports: mainframes are being used more than ever, meaning consumption is naturally on the rise. This is not a bad thing.
However, where consumption is escalating due to inefficient coding, adding unnecessary costs. For example, MIPS costs are increasing on average by 21% year over year, with 40% of survey respondents claiming that consumption is getting out of control. Meanwhile, 88% of respondents using pay structures based on CPU consumption (approximately 42% of those surveyed) think their outsourcer could manage CPU costs better, and 57% of all respondents believe outsourcers do not worry about the efficiency of the applications that they write.
New workloads also are driving costs. For example, 60% of survey respondents believe that the increase in applications like mobile banking are driving higher MIPS usage and creating additional costs. Just think what they’d report when big data analytic applications start kicking in although some of this processing should be offloaded to assist processors.
The Compuware study is interesting and informative. Yes, outsourcers should be pressed to utilize MIPS more efficiently. At a minimum, they should shift workloads to assist processors that have lower cost per MIPS. Similarly, developers should be pressed to boost the efficiency of their code. But this will require an investment in tools to measure and benchmark that code and hire QA staff.
A bigger picture view, however, suggests that focusing just on MIPS is counterproductive. You want to encourage more workloads on the z even if they use more MIPS because the z can run at near 100% utilization and still perform reliably. Higher utilization translates into lower costs per workload. And with the cost per MIPS decreasing with each rev of the zEnterprise the cost per workload keeps improving. Measure, monitor, and benchmark and do whatever else you can to drive efficient operation, but aim to leverage the zEnterprise to the max for your best overall payback.