Check_mk - The Official Documentation

Lets assume you have a network with a ping time (RTA) of 100 ms and connect to ten remote sites. If Multisite would query all sites one after another - you'd have to wait quite a lot for each page to be displayed. This is almost independent of the band width. Even a 2 GBit/s line is not of much use if the RTA is high. Livestatus does not transport much data, anyway.

As of version 1.1.7i2, Multisite brings performance improvements in such situations. It implements two new features:

• Parallelized queries
• Persistant Livestatus connections

In the new implementation Multisite always first sends out a query to all sites and then collects all responses. That way, the network connections to the remote sites can be used in parallel.

In high latency situations the building up of the TCP connection is extremely time consuming. Not only at the connection hand shake, but also later - when the optimal TCP window is determined - several round trips are necessary and add up to substantial times - even if your bandwidth is quite large.

In order to tackle that problem, Multisite now allows for persistent Livestatus connections. These have to be enabled on a per site basis in sites:

sites = {
   # connect to local Nagios
   "local" : {
        "alias" : "Munich"
   },

   # connect to remote site
   "paris": {
        "alias":          "Paris",
        "socket":         "tcp:10.0.0.2:6557",
        "persist":        True,
    },
}

The connections are kept persistent in the Apache processes up to the idle_time which you set in Livestatus (the default is 300 secs). That means - on the other hand - that it might be better, to let Apache create as few processes as possible. This can be done by setting startServers and MinSpareServers to small values in your Apache configuration, e.g.

StartServers       1
MinSpareServers    1