openstack team mailing list archive

[Caitlin Bestler <Caitlin.Bestler@xxxxxxxxxxx>]

Eoghan Glynn wrote:


>> The ultimate goal of distributed dedup is scenario #1. Only the client 
>> software can determine the optimum chunk boundaries,

> Agreed, this is where the maximum savings can be acheived, at the cost of imposing some complexity on the client layer. I'd be interested in using glance as a proof-point for this approach, its one of the topics I 
> intended to discuss at the summit as part of this session:

>  http://summit.openstack.org/sessions/view/78


A lot of the features you are referencing here are good ideas, but I think they belong as part of a Volume Service, rather than Glance.
Glance deals with boot images and their metadata specifically. The optimizations you reference apply to Volume storage in general.

From a distributed Deduplication perspective a Volume is just a collection of deduplicated Chunks, making it pretty much the same as an Object.

But dedup is not the primary issue for volumes, but rather enabling efficient cloning so that then 100 VM's running a given OS can share the 95%
of their boot drive that they have not modified.