Easy-to-use performance tools with a consistent user interface across HP operating systems

Hewlett-Packard Journal, June, 1991 by Rex A. Backman

ALL HP COMPUTER SYSTEM users should be able to use and understand their performance tools. This concept was the guiding principle behind the HP GlancePlus family of diagnostic performance tools created by the performance technology center of HP's Application Support Division. Typically, computer system performance tools have been directed towards the technically advanced user. The focus on tools that required a high degree of operating system knowledge left out novice users. Feedback from our customers regarding computer system diagnostic performance tools subtly addressed this issue by requesting performance tools that were consistent in look and feel across platforms (MPE V, MPE XL, and HP-UX operating systems), were easy to use, and were low in cost. To answer these customer needs, the HP GlancePlus family of diagnostic performance tools was created.

Performance diagnostic tools provide the customer with a means of seeing and understanding the current status of the machine at any time. These tools are typically used to diagnose machines when an abnormal situation is present. Abnormal situations can be a variety of events ranging from poor terminal response time for a specific end user to overall sluggishness in the machine. Whatever the symptom, the customer has deemed something to be out of the ordinary and the performance tool is executed to display real-time data about the machine so that appropriate action can be taken to correct the situation. HP GlancePlus is typically used to determine information such as who is the main CPU consumer, where are the record pointers within a file, who is accessing a certain file, or what disk drive is most heavily used. All of these actions are thpical of the diagnostic quadrant of performance management, and keep the user abreast of the current status of the machine. The performance management quadrant and other types of performance tools are described on page 70.

Background

Historically, many HP performance tools have started as software laboratory prototypes or special tools. These software prototypes were usually trying to address a specific and immediate need in a project-development cycle. For example, new operating systems development work necessitated some form of a monitor to track modifications made in system level algorithms. Upon completion of the project-development cycle, the performance tools would sometimes evolve into a salable customer product. Unfortunately, these customer performance tools still retained the roughness associated with laboratory prototypes. While some effort was spent on improving the prototypes, this work usually concentrated on an improved user interface with no true thought given to the performance data that was displayed. The result of this quick transformation from development tool to customer product was a utility that provided too much data, provided data that was not applicable to the customer's needs, and was too expensive. As a result, performance management on our customer's machines was often relegated to technical experts. This caused an aura of mystique about the performance discipline because the thinking was that if you had to be an expert to use a tool, then the discipline must truly be difficult.

Removing the Mystique

The engineers at the performance technology center recognized, like many others, that performance management was not difficult nor should Hewlett-Packard customers find it so. Several goals were identified at the start of the product cycle to ensure that the mystique about performance management was eliminated. An aggressive development schedule, coupled with frequent and repeated customer feedback sessions, was part of the development methodology used to keep the product team focused on the goal of creating a useful, more informative performance tool for our customers. Customer feedback was the essential checkpoint vehicle used to ensure that the tool was providing valuable and pertinent data (see "Design Prototyuping for HP GlancePlus" on page 69).

This customer feedback loop was important because the HP GlancePlus family is based on a different type of model than previous tools. To help remove the mystique of performance tools, HP GlancePlus incorporates a method of exception-based performance reporting. This approach allows tool designers to limit the data displayed to only those few metrics that are considered interesting. This concept allows the end user to be concerned only with the interesting events on the machine. By focusing only on these interesting events, end users can ascertain what is wrong much faster.

In our diagnostic tools, information can be divided into various hierarchies such as global CPU use, global I/O activity, global memory allocation, and process-specific use. The first three of these hierarchies are condensed into single screens to provide information to the tool user. The process hierarchy presents a challenge because of the possible volume of data. To lower this volume, the exception-based methodology provides only information on processes that have some kind of activity during a specified time interval. Examples of activity for a process include CPU time consumption, disk reads or writes, and terminal transactions. Processes that have engaged in any of these activities during the specified time interval are considered exceptions and are classified as interesting. A process that does not have any activity is of no interest and therefore is not reported.