[mrtg] ext3 sync speed
pavouk at pavouk.org
Thu Jun 30 14:52:46 CEST 2011
> We also had this issue (40,000 targets) for a long time; tried tuning
> commit which helps a bit but not much. Obviously the fastest disks you
> can get help to reduce the bottleneck. RRD 1.4 helps a bit, though
> noatime makes little difference.
I have already applied all these things.
> You get significant savings by making sure MRTG is using daemon mode (as
> opposed to using cron or similar) with RRD 1.4, so that the cfg files are
> cached and the memory-mapped features of RRD1.4 kick in.
I use CRON mode. Do you think, that memory-mapped features are not used
in cron mode?
I have over 1.2GB cached memory. It was worse with older RRDTool versions...
> The main thing that fixed it for us was using rrdcached. However this
> requires you to use the latest version of MRTG, plus the trunk version of
> RRD 1.4 as well as the latest version of Routers2 (if you use this as the
> frontend) and it has a couple of other caveats.
I use 14all.cgi and my own index cgi and search engine.
I don't want to much experiment on this production box. With rrdcached
I will probably wait some time when it stabilize.
> rrdcached is the read/write RRD caching daemon that goes between rrdtool
> and the disk. You can tune it to be more or less aggressive and this has
> a significant impact on disk I/O queues.
Did you know about some filesystem tests? It will be better to use ext4
or btrfs for faster sync?
More information about the mrtg