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

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

Performance algorithms are used by the HP GlancePlus tools to compute metrics for each specified time interval. The differences in software structures across the platforms mentioned above require different access methods to acquire the data. Once acquired, the algorithms are basic ones such as deriving the rate per minute of terminal transactions, disk use percent, and CPU use percent.

The HP GlancePlus tools are designed to be used interactively, or in a real-time mode, with the performance metrics displayed as rates, absolute values, or percentages. To derive these values two time points are required. The difference between the time points is referred to as the interval, and is used as the base for all metrics the user sees. The interval can be as small as five seconds or as large as several minutes. Usually, end users configure the interval between thirty and sixty seconds. As each interval expires, another time point is met, data points are collected, and performance metrics are computed. This refreshing of the performance metrics gives the end user a continuous perspective of what is happening on the machine. Examples of these metrics include system disk rates, which state the number of physical disk I/Os per second, main memory use, which states how much of main memory is being used in megabytes, and global CPU use, which shows the percentage of overall use of the CPU in the most current interval.

Data Sources

Data sources for the HP GlancePlus family differ from platform to platform (see Fig. 4). The underlying differences in the data sources are hidden from customers. However, these separate and unique data sources are the main contributors to the HP GlancePlus performance tools. Each platform has a subsystem associated with the base operating system called the measurement interface. Each measurement interface, regardless of its platform, is composed of two elements: instrumentation and counters. The instrumentation interface is where certain events within the operating systems are recorded. This is accomplished by adding code to specific modules of the operating system kernel that record the activities that the specified modules perform. The added code consists of high-performance statements that simply update the proper counters reflecting the action that has taken place. For example, the completion of a disk read is marked in the disk driver code. Counters are nothing more than accumulators for certain types of operating system events. Counters can be viewed as time stamps signifying the amount of time an event has taken.

Fig. 5 shows the logical map of the MPE XL measurement interface data structure. This structure shows the counters used by the HP GlancePlus/XL tool. While this structure is unique to the MPE XL operating system, it is common for measurement interface structures to organize data in a hierarchy of classes of global, process, and I/O events. In the case of the MPE XL operating system, the counters are organized into these three classes at the highest level. The subclass level represents an intermediate level of organization that provides the ability to identify objects of a similar type. For example, a subclass in the I/O class identifies each logical device present on the machine. Subclasses do not exist in the global class, but they do exist in the process class to identify each process. Each subclass is made up of groups of counters. The counters in each group are logically related by function. An example would be the file system counters for a process. This group of counters contains all the counters associated with file system activities for a specific process. To access a counter, the program must access the proper class (global, process, or I/O), select the proper subclass if applicable, and then access the group that contains the specific counter of interest.