Like all software monitors, SM2 runs under the system to be monitored and thus requires certain resources for its own operation; therefore, strictly speaking, it modifies the system to be monitored. However, this influence is small and can generally be disregarded. For a description of system utilization, see section "System resource utilization by SM2".
Inaccuracies resulting from marginal problems
Ideally, the SM2 activities during monitoring activation and deactivation, during the taking of samples and at the end of a monitoring cycle operand, should take no time at all. This being impossible, certain inaccuracies result. However, this effect is minimal and decreases in direct proportion to the number of actions that have to be performed at any given time (small number of tasks, devices etc. to be monitored).
Inaccuracies resulting from classification
Some SM2 values are gathered on a system-global basis, on a category-specific basis, and/or on a task-specific basis. For category-specific data, SM2 uses the category assignment that is valid at the time the data is collected (sample or event). However, SM2 does not recognize category switches for the TASK monitoring program. This is why comparisons between task data accumulated by category and category-specific SM2 data may lead to interpretation errors.
Monitoring method inaccuracies
Different inaccuracies can occur depending on the monitoring method used:
Event-driven monitoring method
This method supplies very precise data at the cost of increased system workload. Problems can occur only when the duration of events is monitored.
As shown above, the duration of an event (and, if applicable, an activity count) is assigned to the second monitoring cycle even though part of it should be assigned to the first cycle. The relative magnitude of the monitoring error decreases in inverse proportion to the length of the monitoring cycle.
Sample-driven monitoring method
The accuracy of this monitoring method is subject to the laws of statistics.A requirement for the validity of the monitored data is that the samples are independent of the monitored events. SM2 uses the system timers to control sampling by having itself activated at regular intervals.
Hardware interrupts are used for this purpose. However, such an interrupt is not permitted whenever the CPU is in a non-interruptible state. This results in a sampling delay and thus in a certain dependency on system events.Certain unavoidable system activities cause further delays between interrupt acceptance and sampling by SM2. If statistical independence of the samples is assumed, the accuracy of the monitored data depends on the number of samples.
An assessment of the accuracy can be obtained by using confidence intervals (e.g. deviation of not more than 1% in 99% of all cases).
It should be noted that a high sampling rate increases the system workload. Therefore, a long monitoring cycle is preferable to excessively frequent sampling.