IBM Systems Magazine, Mainframe Edition - January/February 2012 - (Page 31)

Tech Corner Programming, systems operations and more Built for Speed How you do what you do when you’re a z196 CPU By Bob Rogers T he latest IBM mainframe processors still execute programs written in the 1960s—and do so thousands of times faster. Of course, this phenomenal speed-up has been fueled by advances in chip technology, but that’s not the whole story. a complete instruction before starting the next, the machine passes a sequence of instructions through a number of stations, each of which performs a part of the overall instruction. These stations are called stages of the pipeline and typically perform activities like instruction fetch, instruction decode, operand location, operand fetch, operations on operands and storing results (see Figure 1, page 32). In this specific decomposition of work, the pipeline has six stages, each taking one CPU cycle to perform its job. Up to six instructions can be concurrently executed in the pipe. In Figure 1, during the cycle that is circled, four different instructions are simultaneously executing in different stages of the pipe. Ideally, each instruction takes six cycles to complete from beginning to end, but the magic of the pipeline is that now the machine can complete an instruction every machine cycle—a speed-up factor of six. Of course, this would only happen under ideal conditions where the pipeline experiences no delays. This is a big caveat ON THE WEB For more about the zEnterprise 196 processor, see Bob Rogers’ Web exclusive companion piece at ibmsystemsmag. com/mainframe/administrator/ performance/rogers_processor/. Processor designers have devised some clever techniques to provide a good chunk of this improvement. Let’s examine some of these interesting tricks. Programmers who know little about internal processor design will be surprised to learn how a zEnterprise* 196 actually executes programs. We’ll start with how a programmer thinks the machine executes a program. A program is a sequence of instructions that operate on data. The programmer imagines that the machine executes the instructions one after the other, completing one before starting the next. This is what the architecture refers to as the conceptual sequence. This conceptual sequence is an illusion—but one the machine must maintain despite the radically different goings-on under the cover. The Pipeline The first technique to violate the programmers’ mental model is called pipelining, which seems to have been adapted from Henry Ford’s idea of the assembly line. Rather than executing JANUARY/FEBRUARY 2 012 31

Table of Contents for the Digital Edition of IBM Systems Magazine, Mainframe Edition - January/February 2012

IBM Systems Magazine, Mainframe Edition - January/February 2012
Table of Contents
Editor’s Desk: Simplicity Itself
IBM Perspective: The Changing Data Center
Insider: Solve the DevOps Challenge With Orchestrated IT
Q&A: Guru Rao Explains How a Better World Begins With Smarter Computing
Case Study: A Comfortable Place: Furniture Brands rests easier with SecureAgent's backup and recovery solution
Cover Story: Smarter By Design: "That's the way to build capacity and hold costs flat," says IBM VP Doug Brown
Feature: Another Dimension: What sets the zEnterprise environment apart?
Tech Corner: How You Do What You Do When You're a z196 CPU
Tips & Techniques: Virtualization Reshapes How Data Is Stored
Solutions: CA Mainframe Application Turner - M4Workbench Eclipse - Virtual Machine Backup
Resources: Advertisers’ Index
Stop Run: Through Three Generations, One Family's Story Shows That While Technology Has Changed, IBM's Fundamental Mission Hasn't
Reference Point - Global events, Education, Resources of Mainframe

IBM Systems Magazine, Mainframe Edition - January/February 2012