Linux on System z is a powerful, stable operating system platform and its popularity is rapidly growing. But like any new operating system, there are specific performance issues that customers must monitor. This is particularly true in a shared, virtualized environment where some of the “rules of thumb” from the standalone, distributed world can actually hurt performance.
This article examines the Linux on System z historical roadmap, including:
• Why many Linux proof-of-concept projects have foundered due to poor expectations and lack of performance management
• The state-of-the-art in Linux on System z performance tuning, with information to debunk some of the performance myths
• Some things Linux on System z sites shouldn’t do to their penguins—and why.
IBM released Linux for System/390 in 1999; it was an IBM “skunkworks” effort, distributed via the Marist College Website rather than through official IBM channels. Linux hadn’t entered the popular consciousness back then.
This first release of Linux for the mainframe caused great excitement throughout the Linux and mainframe worlds. An early adopter claimed to have run more than 41,000 Linux guests on a single VM system, garnering significant press attention. The movement gathered momentum as Linuxcare, Aduva, and BMC released products addressing mainframe Linux provisioning and management.
With the release of the z900 and the Integrated Facility for Linux (IFL), mainframe Linux seemed positioned to take off. IBM pushed it hard, offering “loaner” IFLs to many installations.
The unfortunate result was many failed pilot programs. z/Linux worked, but few knew how to tune it for performance, and the zSeries CPUs of that era weren’t fast enough to compete with cheap Intel processors. So many prospective mainframe Linux sites failed to prove sufficient ROI. For several years, mainframe Linux growth sputtered and stalled; Linuxcare, Aduva, and BMC all exited the market, though IBM and Rocket Software took up some of the slack with new product introductions.