2

I've read in this article that adjusting Ceph's chunk size can improve its performance. How is that set? This document explains how to set the stripe size. Is that what the first article means by "chunk size"?

Don Branson
  • 13,631
  • 10
  • 59
  • 101

1 Answers1

6

Ceph, at its core, doesn't have a "chunk size" (term used in the article, but not on the docs). The RADOS layer stores objects as a whole, without any splitting or striping.

It's the other layers that do striping of their data into several RADOS objects.

If you're using RGW (for S3/Swift compatibility), check the Ceph Object Gateway Config Reference for rgw object stripe size.

For RBD (block device access), check the --stripe-unit argument to the rbd create command.

Finally, for CephFS (mountable filesystem), in the manpage you refer, the --stripe_unit argument is the size of each stripe.

Don Branson
  • 13,631
  • 10
  • 59
  • 101
Javier
  • 60,510
  • 8
  • 78
  • 126
  • 2
    Have you found that changing the stripe size has an impact on IOPS? – Don Branson Feb 24 '15 at 14:42
  • I've tried tweaking stripe-unit/stripe-count/order with no significant impact on performance. This is running Ceph on a local VirtualBox cluster with 2-4 OSDs. Next step is to try this on on a bigger scale. – Don Branson Feb 27 '15 at 14:54