Understanding the sizing charts

Edit online

When the system resources are managed by cpuplugd the amount of CPUs and memory varies over time. The tables list the average memory sizes allocated at the time when the system's CPU load reaches a steady state.

To show the dynamic behavior, the charts depict the individual values over times of interest. However, these charts are not always clear. Figure 1, as an example, shows the number of CPUs over time assigned to all guests.

Figure 1. Number of active CPUs assigned to all guests when managed by cpuplugd
Graph showing the number of active CPUs assigned to all guests when managed by cpuplugd.
It is very hard to see what really happens. With the following simplifications:
  • All systems which are expected never to exceed the load of 1 CPU are withdrawn. In the example, these are lnweb1, lnweb2, lnudb1 and lnudb2.
  • The combo systems are expected to behave the same, therefore only lncombo1 is shown.

The result is shown in Figure 2.

Figure 2. Number of active CPUs assigned to selected guests when managed by cpuplugd
Graph showing the number of active CPUs assigned to selected guests when managed by cpuplugd.
After applying these simplifications you can see that the two WebSphere® servers are handled differently. The chart also shows what happens during the different load phases:
00:00 cpuplugd gets started, the number of CPUs is quickly reduced to 1
00:29 The middleware is started, causing a short load increase to 2 CPUs.
01:41 The workload starts, after a ramp-up phase the number of CPUs assigned to lnwas1 is reduced
11:41 The workload stops, and the number of CPUs assigned to all servers is reduced to one

The charts for memory sizing are optimized in a similar fashion. The optimization rules are explained for each scenario. The memory size considered is the difference of the defined guest size, (5 GB per guest) and the size of the CMM pool, which is reported in /proc/sys/vm/cmm_pages (in pages). Mostly only the size of the CMM pool is shown; but a large pool signifies a small system memory size. The target is to be close to the manually sized setup. If the remaining memory size is smaller than the manually sized setup, it is likely that the performance is negatively impacted.