Determining the capacity of a mainframe isn't always as it seems. One of the main metrics used to calculate it, MSU (million service units), can be imprecise, said Jim Horne, a speaker at last month's Share user group conference.
Horne, a systems programmer at Lowe's, said that using MSU data generated by a mainframe's System Management Facility (SMF) to determine capacity does not always encompass details of which processors go offline and online within 15-minute intervals. Specifically, when a mainframer turns engines on and off with the Intelligent Resource Director (IRD) -- a z/OS-based mainframe tool that dynamically allocates processor resources – he doesn't know how many engines are on at a given time, which means that he doesn't know how fast an individual engine runs. And because some CPU use is reported only in MSU, it's impossible to know how much time a CPU was actually in use.
But Horne said a Java-based tool called zPCR, short for System z Processor Capacity Reference, is helping sort out these capacity-planning hurdles. From IBM's Washington Systems Center, the tool is not yet an official product of the company and is offered for free as a download on an as-is basis. It is a Windows-based software program designed to provide capacity-planning insight for mainframe processors running various workloads under z/OS, z/VM and Linux, and is based on data from the Large Systems Performance Reference (LSPR), which is IBM's measurement of processor capacity based on processor design and performance on various workloads.
According to Horne, zPCR "is indispensable right now if you're going to do real capacity planning." That said, zPCR has its issues, including the lack of an application programming interface -- which Horne says makes it onerous to extract data – as well as the absence of an automatic updating process. But because it's a tool still in development and not an official IBM product, support isn't as immediate as it is with other software.
"It's a great tool, it's invaluable," Horne said. "But it's got problems too."
The mainframe capacity measurement problem
Horne detailed his concerns about the traditional ways of measuring mainframe capacity. MSUs are a measurement of how much processing work a mainframe can perform in one hour, and are used mainly by IBM and third-party software companies to determine licensing costs. Mainframe shops also use them to determine a box's capacity and to report performance numbers.
Meanwhile, SMF is part of the z/OS operating system that provides a way to write out certain aspects of a mainframe's activity to a file. Some of these activities include processor utilization, I/O performance and error conditions. Another feature of z/OS called Resource Management Facility (RMF) uses SMF files to provide performance and capacity measurements of processors, memory, virtual storage and other mainframe components, such as MSU.
Horne said that two common processor capacity and performance SMF records, type 70 and type 72, can lead to imprecise RMF measurements.
"RMF tells us what happens in 15-minute intervals, so we don't know when engines are going on and offline," Horne said, later adding that "reporting may already be introducing errors that you may not be aware of."
According to Horne, the SMF type 70 record reports software capacity and the type 72 record reports the hardware capacity for a particular mainframe logical partition (LPAR), but neither reports hardware capacity for an entire machine (unless you run only one LPAR).
By using the zPCR tool in concert with analyzing SMF records, Horne said he's been able to get a better grasp on Lowe's mainframe capacity and processor use.