loadavg-based

The loadavg-based configuration is one of the rules used for managing CPUs with the cpuplugd daemon.

The full configuration file is listed in cpuplugd configuration files.

The lines of interest are:

HOTPLUG="(loadavg > onumcpus + 0.75) & (idle < 10.0)"
HOTUNPLUG="(loadavg < onumcpus - 0.25) | (idle > 50)"

These lines implement the following rules:

The system plugs CPUs when there are both more runnable processes and threads than active CPUs and the system is less than 10% idle
The system removes CPUs when either the amount of runnable processes and threads is 25% below the number of active CPUs or the system is more than 50% idle

Figure 1 shows the effect of the CPU management rules over time:

Graph showing the number of active CPUs over time when managed by cpuplugd based on loadavg value — Figure 1. Number of active CPUs over time when managed by cpuplugd based on loadavg value

Table 1. Throughput and average CPU load when managed by cpuplugd based on loadavg value
Throughput and average CPU load
Configuration	TPS*	Relative CPU load*
loadavg-based	88%	84%
*100% is the manual-sized run

Observation

The number of CPUs is lower than manually sized for most of the time. The combo-system is frequently reduced to 1 CPU for a short time. The throughput is significantly reduced.

Conclusion

The value of loadavg determines the amount of runnable processes or threads averaged over a certain time period, for example, one 1 minute. It seems that this value changes very slowly and results in a system running short on CPUs. The guests running a WebSphere® application server are always highly utilized, but the rules do not add the required CPUs, which leads to the observed reduction in throughput.

This configuration is probably useful when trying to restrict the addition of CPUs and to accept that the guests run CPU-constrained with the corresponding impact on throughput, for example in an environment with a very high level of CPU overcommitment.