From collectd Wiki

The RRDCacheD plugin connects to the “RRD caching daemon”, rrdcached and submits updates for RRD files to that daemon. rrdcached is based on the RRDtool plugin of collectd and basically uses the same principles that are described in “Inside the RRDtool plugin”.

Having the cache in a separate process is some more work to set up, but has nice benefits, especially for big setups: You can restart collectd without flushing the cache, have the cache on a remote server and, of course, applications other than collectd can make use of it.

rrdcached has been included in the RRDtool distribution since version 1.4.

Synopsis

<Plugin "rrdcached">
  DaemonAddress "unix:/var/run/rrdcached.sock"
  DataDir "/var/lib/collectd/rrd"
  CreateFiles true
</Plugin>

Note that if you were using rrdtool previously and are now turning on rrdcached support, then you will need to recreate DataDir so that it's empty and can be initialized by rrdcached. The rrdcached plugin will fail to read from an existing non-rrdcached-initialized storage location.

Example graphs

Number of write operations per second.

Values written to disk per second. Since many values are written in one write operation, the number of write operations per second is much lower. As you can see if you compare this graph with the one above, there are in average 490 data sets in one write operation.

Number of update commands received per second.

Bytes per second written to the journal.

Length of the update queue.

Number of nodes (RRD files) in the binary search tree of the caching daemon.

Depth of the binary search tree used by the daemon. This is basically the number of comparisons needed to find any file in the tree.

See also

• Inside the RRDtool plugin
• RRDtool plugin

Dependencies

• librrd (≧ 1.4. Version 1.3 is not recent enough!)