graphite-dev team mailing list archive

[chrismd <question170794@xxxxxxxxxxxxxxxxxxxxx>]

Question #170794 on Graphite changed:
https://answers.launchpad.net/graphite/+question/170794

chrismd proposed the following answer:
The retention schemes really only affect the size of the wsp files and
so cutting it back will only save you disk space not throughput. Slight
caveat to that statement actually, since the file sizes are larger
creates are more expensive with larger retentions but that only matters
when creates are happening and is mitigated by MAX_CREATES_PER_MINUTE
anyways.

Kevin's suggestions are right on target (use ext4 and raid10 for sure).
Though I had no hand in the hardware setup (so unfortunately I cannot
provide much detail), the most efficient Graphite system I've worked
with uses an array of SSDs. One of the key characteristics of solid-
state is reduced seek time, which is the single biggest performance
factor for Graphite's I/O workload. They've got a single box pushing
750k metrics/min and they're one of the few Graphite users who worry
about CPU bottlenecks more often than I/O. If I were to build a Graphite
system today, I'd do it with an array of SSDs. That said, I'd need to do
some research and experimentation to determine the specifics for an
ideal system based on SSD. The next biggest Graphite system I've worked
on that I do know more details on uses ext4 and RAID10 with 8 10k rpm
spindles.

I hope that helps. If you're able to post any conclusions from your
testing it would be very much appreciated.

-- 
You received this question notification because you are a member of
graphite-dev, which is an answer contact for Graphite.