Impact of sensor configuration

Sensor configuration has an impact on pmcollector memory consumption.

When you enable a sensor, there is an impact on the collector memory and disk consumption and the query processing time. The impact depends on the system configuration and the number of keys and metrics that are provided by the sensor. For example, sensor NFSIO keys and metrics scales by the number of protocol nodes (cesNodes) and the number of NFS exports.

A high number of NFS exports or CES nodes has a large impact on the pmcollector memory consumption. In addition, the sensor scheduling frequency period setting has an impact on how many metric values are sent to the collector. A lesser period value means that metric values are sent more often to the collector that increases pmcollector memory consumption.

The following table lists and explains the impact of enabling certain sensors and how the impact scales.

Table 1. Sensor with high impacts
Sensor name	Scaling factor	Default period setting	Comment
NFSIO	Number of cesNodes * number of NFS exports	10	The key consists of node, sensor, export, and `nfs_ver` (NFSv3, NFSv4), 12 metric values per key. For more information, see NFS metrics.
GPFSDisk	Number of NSD server nodes * number of disks	0 (disabled)	The key consists of node, sensor, `gpfs_cluster_name`, `gpfs_fs_name`, `gpfs_disk_name`, 16 metric values per key. For more information, see GPFS metrics. Warning: Enable the sensor on demand only and restrict to the nodes of interest.
GPFSAFMFSET	Number of AFM gateway nodes * number of file sets	10	The key consists of node, sensor, `gpfs_fs_name`, `gpfs_fset_name`. More than 130 metric values per key. For more information, see GPFS metrics.

Recommendations

System entities that do not exist anymore leave their key signature and their historical data in the collector database.
For example, if you delete a node, file system, file set, disk, and other resources from the cluster, the key and data of the removed entity remains within the collector database. The same is valid if you rename an entity (for example, rename a node). If your system entities are often short-living and for temporary usage only, such entities might cause a high impact on the pmcollector memory consumption.
You must check your system for such expired keys from time to time and delete the expired keys by running the following commands to list the expired keys first and delete afterward.
```
mmperfm query --list=expiredKeys
```
```
mmperfmon delete --expiredKeys
```
For more information about command usage, see mmperfmon command.
Note: The historical data of the removed entities is not removed from the database automatically with immediate effect because it is possible that the customer might need to query the corresponding historical data later.