Do you have guidelines or averages of bandwidth requirements for clustering? For example, an average of 10kbps per 1,000 sensors?
This article applies as of PRTG 22
Bandwidth usage in a cluster
- A cluster only transfers changes. If all sensors are up and running, and no objects are being edited, no data is transferred between the cluster nodes.
- There is a Keep alive message from each cluster node every 10 seconds, which does not scale with the sensor count, however.
- On startup or if the cluster nodes are out of sync, there can be a transfer spike when the entire configuration is sent to the other cluster nodes (in a compressed form). You can have a look at the size of your non-compressed configuration file, but as this varies from user to user, it is impossible to name an exact number here.
- If sensors change to the Up or Down status, the status is broadcast to all other cluster nodes. The traffic caused by this depends on the number of cluster nodes.
- If the PRTG core server on one of the cluster nodes is updated, the setup file (about 36 MB) is transferred to the other cluster nodes to update them automatically.
- When the reviewing of monitoring data includes different cluster nodes, the required data (states, graphs, tables) is transferred from the other cluster nodes to the cluster node you are logged in to. Graphs in particular for which All cluster nodes was selected can cause traffic peaks. This affects reports with an All cluster nodes setting, too.
- In a cluster, every cluster node independently monitors the devices added to the cluster probe. Each of these devices sends monitoring data to every single cluster node. Depending on the number of cluster nodes used, this causes extra traffic.
Note: Especially when many WMI sensors are used, this can lead to extended traffic on the hosts.
What impact should we expect to see in a two node clustered setup on our monitored devices. Are we doubling the amount of SNMP operations that devices must performed, because they are being checked by each clustered node? What considerations has Paessler made for this?
Yes, when running a cluster configuration, the SNMP queries are performed by both servers thus doubling the amount of SNMP traffic but this traffic is rather negligible and will only occur for those sensors which are underneath the Cluster Probe.