Response Time

While you can rely on end-users to tell you that there is a performance problem, you will need some figures which allow you to quantify it, so that the effect of changes made can be accurately measured.

Degrading response time for simple interactive transactions like scrolling in an XEDIT session is usually the main reason for user complaints. Actual interactive (or trivial) response time would, therefore, be an ideal performance indicator. Unfortunately, no information is available in system control blocks which would allow the calculation of this end-user response time, which is why it cannot be provided by the Performance Toolkit either. In the absence of real response time, it displays the next best internal response time indicators available in VM:
  • The average internal response time for trivial transactions (transactions requiring just a single class 1 slice). The values are shown under the heading 'TR-T'.
  • The average internal response time for non-trivial transactions. The value is shown under the heading 'NT-T'.
  • The class 1 elapsed time slice. The value is shown under the heading 'C1ES'. The VM scheduler adjusts this value dynamically to keep the appropriate distribution of transactions in the various scheduler user classes. The class 1 elapsed time slice is a good indicator of bottlenecks as wait times, CPU, I/O, and paging are factors in its computation.
Neither the TR-T value nor C1ES are really equivalent to the actual interactive response time experienced by end users (neither include, for example, the time for data transfer between the user's terminal and the CPU), but you can expect both of them to be roughly proportional to that part of the response time which is directly influenced by the system. Both can be used, therefore, as indicators of general system responsiveness to trivial user commands. We will mainly refer to the C1ES value in the following examples.

Look at the corresponding 'non-trivial response time' values under heading NT-T if you need an indicator of system performance for non-trivial commands or batch jobs.

For VM systems made up of mostly guest operating systems, response time values are not indicative of the response times of commands with the guest. From VM's view, most guests appear to issue a few very large transactions. These large transactions may represent many smaller commands running inside the guest.