Number of supported devices

Hi All,
Using the raw edition how many devices can a single slave server support? Also does the performance data gathered on a device reside on the slave or on the master server?

TIA

It depends on the specs your slave is running on. You can get more information in the documentation:
https://checkmk.com/cms_cmc.html#The%20(CMK)%20micro%20core

The performance data is stored on the slave and get’s only transferred automatically in CEE. But you can write a cronjob to just rsync the data (eg. hw/sw inventory) over to your master, it’s also explained in the docs.

For the graphs you can configure a reverseproxy connection to your slaves.

As an example we have a building with over 8000 switches in it. With a slave with a lot of cpu and memory can all those be monitored by the one slave.

Performance data stays on the slave all the time. Only HW/SW inventory data is transferred to the master.

For such a high amount of devices (SNMP devices need more CPU than agent devices) i would split the devices on more than one instance.
RAW edition has a “soft” limit per instance i would say around 1k hosts (this is no real hard limit).
What you can do is have a very heavy iron as your monitoring server and run several instances on this one machine. If you have enough resources it is also possible to handle 8k hosts on one machine.

1 Like

I really appreciate your help Andreas. One last question, we have some monster nexus devices with hundreds of interfaces. Is it possible to only identify specific interfaces on a large device like this ? Say only the TE interfaces, and not all the rest of them. I’m thinking if we only look at specific interfaces like uplink interfaces then that would decrease the amount of metrics being polled.

Checkout the following blog post.

3 Likes

Yes that is possible with this rule:
(Interface and Switch Port Discovery)

Per default only ports in state UP will be discovered but you can define with this rule the state or type of port you want. :slight_smile:

1 Like

This is only possible with the enterprise edition. There you have a rule “Bulk walk: Limit SNMP OID ranges”. Only with this rule you can define for the interface check only to fetch the first 10 interfaces or interface 20 to 30 and so on.
All the other rules mentioned before can only be used to limit what you see but not what you get with your snmpwalk.
I know the problem with the very big core switches. One nice idea from my side would be an discovery where you can select the interfaces you want and CMK will build then the rule for the limits by itself :slight_smile:

Thank you for pointing us to a possible solution. After reading through the “four rules” doc, it appears we may be facing an larger issue . We have thousands of devices, mainly only distribution devices have comments in the if alias’s section but not all. Different sites kind of do their own thing so there’s little uniformity. How can we stipulate from the Description field information only monitor active FortyGigabitEthernet, HundredGigabitEthernet ports for alerting and ignore the rest?

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.