[one-users] Copy image on Ceph to file-based datastore
Stuart Longland
stuartl at vrt.com.au
Sun Mar 23 13:42:37 PDT 2014
Hi Jaime,
On 21/03/14 19:57, Jaime Melis wrote:
> Have you benchmarked Online write-through caching / Online write-back
> / Offline write-back ?
Benchmarked? I haven't even written the code yet.
At the moment without caching I'm getting about 120MB/sec in my VMs when
using virtio storage. I've been told by one of the would-be users that
this is "unusably slow".
Sébastian Han did some work with RBD+FlashCache here:
http://www.sebastien-han.fr/blog/2012/11/15/make-your-rbd-fly-with-flashcache/
> Are you proposing we should allow the three options or just stick to one?
>
> Anyways, this sounds like an amazing addon :)
Well, I can see cases where all three are useful. Production data
workloads where data integrity is crucial, you'll want write-through
caching.
Speed-critical work loads, one of the write-back modes would be faster.
Online write-back if you need live migration or don't have the space
for offline write-back.
> http://opennebula.org/addons/create/
>
> It'd be nice if in the machine template where you specify individual
> disks, if you could state what caching mode to use. Alternatives would
> be to specify it in the image template or (least favourable) in the
> datastore template.
>
>
> If I understood correctly, you can use the CACHE option here:
> http://docs.opennebula.org/4.4/user/references/template.html#persistent-and-clone-disks
>
> although it might need a change in order to be able to specify the
> target cache. How do you think the interface should be?
That looks more like how libvirt handles the cache to the back-end
device. Then again, we're pretty much taking over whatever caching
libvirt does and supplanting our own, so maybe it's appropriate to
hijack that option and just tell libvirt to disable its cache.
--
Stuart Longland
Contractor
_ ___
\ /|_) | T: +61 7 3535 9619
\/ | \ | 38b Douglas Street F: +61 7 3535 9699
SYSTEMS Milton QLD 4064 http://www.vrt.com.au
More information about the Users
mailing list