Information Technology Reference
In-Depth Information
7.5.3 Performance Measurements
Once criteria and algorithms have been defined performance runs should be sched-
uled during acceptance tests. This may happen in parallel with the usual test
scenarios or during special time windows reserved for this purpose. Some times
the test of a batch function can be used for performance analysis at the same time it
is tested against possible errors.
7.5.4 Monitoring
Dynamic data are obtained under load as a function of time—generally under real
or simulated production conditions. Other possible configurations include:
Benchmark configurations
An almost empty machine to test specific processes with the aim for later
calibrations and scaling.
To capture dynamical performance data the market or suppliers of operating
systems offer special monitors, which collect these data and visualise them on
screen or make them available in file reports. Depending on the problem under
consideration and on the application environment such monitors will run hours,
days or even weeks either during production or in a simulation environment by
using specifically developed test cases. The most important information to be
collected is the following:
Application programs currently in use,
Distribution of user frequency over a lengthy period of time,
Account statistics,
Processor load,
Main memory usage,
Memory management,
Number of running processes,
Frequency of file accesses (open table),
Overheads and interrupts,
Waiting queues,
System table usage,
I/Os,
Swap rate.
These data can be obtained either as snap shots of specific performance situa-
tions or as a dynamic collection over longer time intervals. Monitors can be
supplemented by log files for system events against the backdrop of the production
schedule. Furthermore there are monitors, which register system interrupts and their
internal transaction times along with addresses in main memory. This detailed
