*6.2. Disk-Related Metrics*

In the evaluation of disk-related metrics, data were collected for the following metrics: READ, which represents the amount of kilobytes per second that the task took to be read; WRITE, which is the amount of kilobytes per second that the task sent to be written to the disk; and finally CANCELLED, which is the amount of kilobytes per second whose disk writing was canceled by the task, that can also occur when the task truncates some dirty page cache. All these metrics were monitored by the Pidstat tool in both Minikube and K3S.

In Figure 12, the WRITE and CANCELLED metrics have their behavior unchanged throughout the experiment, always walking close to 0 KB/s. Although, the READ metric had a distinct behavior, holding the same value throughout a single cycle of the cluster stress, and presenting a linear growth among cycles until the fourth execution cycle, being interrupted abruptly when reaching about 4,000,000 KB/s due to the limiting factor of the Minikube environment. Such behavior may be indicative of the software aging phenomenon in this environment.

**Figure 12.** Disk-related metrics in Minikube.

In Figure 13, the WRITE and CANCELLED metrics also remain close to 0 KB/s throughout the experiment, similar to the execution in the Minikube environment. However, the behavior is different in the READ metric, which was not interrupted abruptly and had a linear growth from one cycle to another until the end of the experiment execution in K3S. It is important to mention that K3S presented smaller values of bytes read per second than Minikube, which might have prevented it from the abrupt fall observed there.

**Figure 13.** Disk usage in K3S.
