My Nagios boxes run pnp4nagios+npcd to record plug-in performance data to RRD files. On one system I enabled rrdcached on, a little over 9,000 files are updated every five minutes. I saw a 30% decrease in disk write operations/second, all other things held constant. There wasn’t a noticeable difference in bytes/second, which I guess is somewhat expected. I’d be curious to see what happens on a box with 100k RRDs, but I don’t have anything that big.
On its own, npcd already does a pretty good job of distributing writes out over time so all 9,000 writes don’t get committed at the same time, at an average of 436 ops/sec. Having said that, running with -w 300 -z 300 should be basically the same thing, but it was a 15% improvement at 357 ops/sec. Going further to -w 900 -z 300 to flush every fifteen minutes brought this down to 301 ops/second.
