Medallia and Redhat Talk Docker+Ceph

Posted Jul 2, 2015 16:52 UTC (Thu) by jberkus (guest, #55561)
In reply to: Medallia and Redhat Talk Docker+Ceph by kreide
Parent article: News and updates from DockerCon 2015

Hey, thanks for the link.

There was a presentation on Ceph at CoreOSFest, too, but I was kinda disappointed in it. I'm hoping that you can answer a couple questions that speaker didn't have time to take:

* What about running Ceph itself in containers? Given the complicated setup of a new Ceph cluster, it seems like containerization would really make it easier to deploy.

* You mention databases in your slides. But clustered object stores, as a rule, really really suck for small frequent writes. How does Ceph handle this?

Of course, if you can do a San Francisco Postgres or Docker meetup, then I'd be glad to watch you present and answer my questions in person.

to post comments

Medallia and Redhat Talk Docker+Ceph

Posted Jul 3, 2015 2:14 UTC (Fri) by thorvald (guest, #103394) [Link] (4 responses)

We (also Medallia) have Ceph monitors in docker; we want the easy relocation without updating ceph configs. For the OSDs, we run them directly on bare metal; we have their installation provisioned as part of the installation automation.

We run PostgreSQL on top of the rbd block devices that Ceph exports, and Ceph does partial block updates really well; we're seeing "all-local-SSD-array" equivalent performance on database transaction benchmarks. Our Ceph cluster is 64 nodes, each with 8 SSDs though, so it's not an entirely fair comparison.

Medallia and Redhat Talk Docker+Ceph

Posted Jul 4, 2015 6:05 UTC (Sat) by jberkus (guest, #55561) [Link] (3 responses)

Huh! Do you have any online numbers for that cluster?

And, if you can do a Postgres meetup in the Bay Area, I think we'd be interested ...

Medallia and Redhat Talk Docker+Ceph

Posted Jul 6, 2015 17:39 UTC (Mon) by drag (guest, #31333) [Link] (1 responses)

I suspect that unless you have 10GbE for your ceph storage network running any sort of RDBS on it is going to be pretty painful.

Medallia and Redhat Talk Docker+Ceph

Posted Jul 7, 2015 20:38 UTC (Tue) by thorvald (guest, #103394) [Link]

We do 40GbE end-to-end, so we peek out around ~4.5 GByte/sec of "disk" I/O.

Medallia and Redhat Talk Docker+Ceph

Posted Jul 7, 2015 20:50 UTC (Tue) by thorvald (guest, #103394) [Link]

I'll go rerun our benchmarks, my notes here say "~6000 tps", but doesn't include the parameters for filesystem or versions, nor which transaction is actually benchmarked. My bad for not taking proper notes.

We'd love to present this at the PostgreSQL meetup; what's the best way to get that going?