Update README.md

Author: pleimer

README.md

@@ -1,8 +1,8 @@
 # collectd-libpod-stats
 Collectd plugin for gathering resource usage statistics from containers created with the [libpod library](https://github.com/containers/libpod). 
 
-## Current functionality
-This plugin is being developed with the new [go-collectd](https://github.com/collectd/go-collectd) library which has not yet released version 1. As a result, the functionality of this project is bound by the ever changing abilities of the library. This means, many features that would be expected from a plugin cannot yet be implemented until the library makes it possible to do so.
+## Functionality
+This plugin is being developed with the new [go-collectd](https://github.com/collectd/go-collectd) library which has not yet released version 1. As a result, the functionality of this project is bound by the ever changing abilities of the library. Many features that would be expected from a plugin cannot yet be implemented until the library makes it possible to do so.
 
 As of now, this plugin reports the following for each container on a node:
 
@@ -14,6 +14,13 @@ memory usage | bytes
 
 There is no way to configure these options.
 
+## Notes on memory usage calculations
+Calculating memory usage on linux systems is hairy and differs based on projects. This plugin replicates the memory calculation that the podman tool uses, that is, the value of `memory.usage_in_bytes` in the case of cgroups v1 or `memory.current` for cgroups v2. This value includes both RSS and CACHE memory values and thus is different to what the `free` tool reports as used memory by reading `/proc/meminfo`.
+
+Kubernetes calulates the memory usage as `memory.usage_in_bytes - total_inactive_file` [(source)](https://github.com/kubernetes/kubernetes/blob/dde6e8e7465468c32642659cb708a5cc922add64/test/e2e/node/node_problem_detector.go#L242). Kubernetes uses this value to make [eviction decisions](https://kubernetes.io/docs/tasks/administer-cluster/out-of-resource/#eviction-signals) and thus this value is important. However, k8s provides its own means of monitoring pod memory and should be used instead of this plugin.
+
+Ultimately, this plugin uses the stated values because the linux cgroup oom_killer uses this value to destroy processes that exceed the hard limit.
+
 ## Build from source
 
 Collectd library files must be made available to the golang compiler to build this plugin. Clone the [collectd project](https://github.com/collectd/collectd).