| As illustrated in Fig. 5.6, a measurement process involves three major steps: measurement specification, system instrumentation, and data analysis. Specify measurements. In this step, the performance variables to be measured are selected. For example, suppose that the behavior of a specific virtual memory policy is of interest. In this case, performance variables such as the page fault rate, the throughput of paging disks,and the average number of jobs competing for memory space, are required by the system model. Instrument and gather data. After selecting the variables to be observed, the system is instrumented to gather the specified measurement data. This involves configuring the tools to measure the specified variables during the observation period and recording the required information. A computer system can be viewed as a series of layers that create an environment for the execution of application programs, as shown in Fig. 5.6. Depending upon the variables selected, several measurement tools may be required at various layers. For instance, if transaction service demands are required, measurement tools are needed at both the operating system level and at the transaction/DBMS level. Analyze and transform data. Measurement tools gather potentially huge amounts of raw data, corresponding to a detailed observation log of system activities. Usually, raw data specify time intervals, event counts, transaction IDs, bytes transferred, and resources consumed. To be useful, these data items have to be mapped to their corresponding logical functions. That is, these bulky data items must be analyzed and transformed into meaningful information. For instance, information recorded by software measurement tools might include a record for each process that starts or completes during the observation period. These records must be manipulated to yield useful results, such as average execution time, number of processes executed, and device utilizations.
Figure 5.6. A representation of the measurement process.  |
