IBM SG24-4763-00 manual Memory Bottlenecks, Measuring Multiprocessor Load

This soft copy for use by IBM employees only.

Figure 112. CPU Utilization with TME 10 NetFinity

8.4.1.1 Measuring Multiprocessor Load

In a multiprocessor server, the best way to measure the overall load experienced by the combination of CPUs is to check the following object (Figure 113):

Object: System

Counter: % Total Processor Time

Figure 113. Multiprocessor Load Measurement

If this value is consistently high (80% or more) and disk as well as network counter values are low, then an upgrade to the processor may be the solution.

Note: Refer to 8.7, ªNetwork Components Bottleneckº on page142 for more details on network object monitoring.

With TME 10 NetFinity, you will see that as soon as you have multiple CPUs installed, you will have monitors available for each CPU. Refer to 6.6, ªSystem Monitorº on page 116 for more information.

8.5 Memory Bottlenecks

Windows NT has the ability to self-optimize certain aspects of the system. This self-optimization or tuning is focused on memory caching and pagefile administration. The tuning is achieved by using memory caching and balancing it by growing the size of the pagefile in anticipation of the load demand that the system expects. The memory management pre-emptively adjusts the amount of caching memory available to best suit the operating conditions.

This is why, on a system that is heavily used, the amount of unused memory seems to be high. The system is trying to buffer the ultra-high demand peaks

136PC Server and Windows NT Integration Guide