Before discussing bottleneck analysis in detail we should recall the basic problems involved in meeting performance expectations:
Performance expectations of the individual user
The performance expectations of the individual user are temporal in nature (response time, elapsed or dwell time). The time taken to process the user's requests to the IT system has a direct effect on his/her work pattern and hence on his/her “productivity”.
Processing time comprises service time and the time spent waiting for the relevant resources. While service time reflects the immediate resource requirements, wait time depends on the workload placed on the resources. Therefore, the lighter the resource workload, the easier it is to satisfy the performance expectations of the user.
Performance expectations of the organization running the server
For economic reasons, the performance expectations of the organization running the server are directed towards attaining maximum “system throughput” (high transaction or job throughput rate).
Therefore, to meet the performance expectations of both the individual user and the systems support, it is always necessary to reach a compromise.
Understandably, investment planners also demand the fullest possible utilization of resources. However, in actual practice, various standard values have emerged for using resource capacity. These values should never be exceeded unless absolutely necessary. In most cases they do not represent technical limits. If exceeded, they lead to inefficient utilization of the IT system and make it more difficult to meet the performance expectations of the individual users.
Therefore, before carrying out measurements, it is essential to clarify which performance expectations are not being met:
System-oriented performance expectations (the system throughput leaves something to be desired, i.e. the performance of the entire IT system is unsatisfactory).
Indications include:
excessively low transaction rate (as an immediate offshoot of generally poor response time behavior)
excessively low throughput rate (dwell time for jobs generally too long).
User-oriented performance expectations (the performance expectations of individual users are not being met).
Indications include:
excessively long response time for certain types of transaction
excessive dwell time for particular jobs.
If the performance problem is system-oriented, the causes most likely lie in an overload on one or more resources.
If user-oriented, it is essential to find out the reasons why delays arise when handling special load requests.