Previous Table of Contents Next


Finally, most accounting logs contain no system-level information, such as queue lengths or device utilizations. The elapsed time includes service as well as queueing time at various resources along with the user think times. It is not possible to separate the effects of queueing (system-level phenomenon) from user service demands. Many programs, including most editors, frequently wait for user input throughout execution. The accounting logs do not record the time spent waiting for user inputs. Thus, there is no way to distinguish between this delay and queueing delay.

In spite of these disadvantages, accounting logs are used extensively for performance analysis.

TABLE 8.1 Typical Data Recorded by Accounting Utilities

Name of the program
Program start time: date and time of the day
Program end time: date and time of the day
CPU time used by the program
Number of disk writes
Total disk write bytes
Number of disk reads
Total disk read bytes
Number of terminal writes
Total terminal write bytes
Number of terminal reads
Total terminal read bytes
Number of page-read I/O’s to the paging device
Number of pages read from the paging device
Number of page faults

8.4 ANALYSIS AND INTERPRETATION OF ACCOUNTING LOG DATA

The data recorded in the logs consists primarily of resource usage: CPU time, elapsed time, number and total size of disk I/O, paging I/O’s, terminal I/O’s, network I/O’s, and so forth. The data may consist of system wide resource usage, resources used by each program, or resources used by each user session. In addition, abnormal conditions such as security violations, abnormal job terminations, system restarts, downtimes, and device errors are also recorded in these logs.

Table 8.1 lists some of the workload parameters recorded in the accounting logs. In the table, the page faults and page-read I/O’s are listed separately because in many operating systems, pages are prefetched and every page fault does not require an access to the paging device. It is possible to derive a number of other data items from these that can provide useful information about the workload. These data items and their usages are described in this section. This discussion assumes that the accounting log records data for each program activation. In other words, the workload unit or workload components are programs. The discussion is general and applies to other workload units such as users, jobs, sites, and applications.

For each program, many different data items can be calculated. In particular, all resource consumptions can be expressed in four ways:

  Per activation
  Percentage of total resources consumed by all programs
  Per-second resource consumption rate
  Per-CPU-second resource consumption

These derived quantities can then be interpreted as follows:

Per-activation data is the average resource consumption per activation of the program. This information is useful for performance analysts in modeling program behavior and in constructing synthetic workloads.

Percentage of total resource consumption is the resource consumed by all activations of a particular program, expressed as a percentage of the total resources consumed by all activations of all programs. For example, if the percentage of total disk I/O’s for program A is 10%, it indicates that 10% of all disk I/O’s on the system can be attributed to program A. To developers and product managers, the percentage of total data items shows the opportunity that different programs offer for reducing consumption of that resource. If the programs are sorted in order of decreasing percentage of total consumption of a particular resource, the programs on the top of the list will have a high impact and the programs on the bottom of the list will have little impact. Thus, if program A consumes only 0.01% of the total CPU, it may not be worthwhile to devote manpower to optimize this program for better CPU efficiency. Since elapsed time is also people time, programs with a high percentage of total elapsed time may provide high potential for improvements in worker productivity.


Previous Table of Contents Next

Copyright © John Wiley & Sons, Inc.