alternate backend storage

BTW Ceres is used in PROD, especially in big installations, e.g. in Yandex, they even have own branch with ceres specific fixes - https://github.com/dkulikovsky. Ceres gave quite big gain for disk space, but not so big gain for IOPS - SSD still required for big load.
You can also try cianite - https://github.com/pyr/cyanite + https://github.com/brutasse/graphite-cyanite - but not sure if whole Cassandra cluster is cheaper than bigger SSD...

Ah, and also http://influxdb.com/ exists. Check out @dieterbe's articles about it - http://dieter.plaetinck.be/post/on-graphite-whisper-and-influxdb/ and http://dieter.plaetinck.be/post/influxdb-as-graphite-backend-part2/

Sure, if you like a TSDB that changes its clustering design every 3 months. *cough*

PPS: OpenTSDB storage plugin for graphite - https://github.com/mikebryant/graphite-opentsdb-finder
http://opentsdb.net/ is required, of course - and it's much harder (from operational point of view) than Cassandra IMO :)

I have been specifically waiting for influxDB because it holds alot of promise but it has yet to mature. I had version 8.x installed and I had the graphite webapp reading from it but there are numerous changes in version 9.x. I think it will be a leader when a prod version is released.

There are also the concerns of having the graphite webapp read from the alternate backend since many use a tag/label and values. The functions on the graphite webapp still seem to be unparalleled.

I know this post is a loaded question but I figure I would at least be humored. The link below seems to match my concern.
https://wikitech.wikimedia.org/wiki/Graphite/Scaling

The time-series database space will be awesome in a year but what do we do now.
http://prometheus.io/
OpenTSDB
influxDB

I love (sarcastic) theoretical discussions like this which completely overlook the fact that numerous companies use Graphite at significant scale in production.

I hope I didn't come across as sarcastic because I sincerely was hoping to find ideas for a short term solution until undertaking a larger project. As I mentioned in my previous comment, I also knew it would be a loaded question. Not trying to offend anyone. Our current graphite infrastructure handles 1.5 million metrics/min but I'm pausing and looking around before scaling it yet more.

Not at all, it was my own personal observation / frustration towards the community at large that want a magic pill for time-series storage. This thread just seemed like as good a place as any to release a brief rant. ;-)

Thanks for the brainstorming/comment all.