If you decide to run collectd on your system, don’t do what I did: I ran it for 8 months without caching enabled. Basically, that means that not only was collectd monitoring load, it was generating a huge amount of load itself — way more than anything else running on the system.
Case in point:
The “no caching” bit is what it’s been running like for 8 months, using this config in /etc/collectd/collectd.conf:
<Plugin rrdtool>
DataDir "/var/lib/collectd/rrd"
</Plugin>
To more than half my idle CPU usage, and get my load averages back down to near 0.00, all I had to do was add:
<Plugin rrdtool>
DataDir "/var/lib/collectd/rrd"
CacheTimeout 120
CacheFlush 900
</Plugin>
I don’t get why they don’t ship it like that by default. Enable caching. Your server will thank you. Thanks to the folks on #slug who held my hand while fixing this. 
