[Sheepdog] [PATCH v5 4/8] sheep: teach sheep to use object cache
namei.unix at gmail.com
Tue Mar 27 08:07:25 CEST 2012
On 03/27/2012 12:01 PM, huxinwei wrote:
> We can, if these VMs are actually cloned from the same snapshot ;)
> BTW: I'm not aware that you are planning data dedup already for farm.
> That'll be really awesome ;)
> However, 4M is far too big for effective deduplication, IMHO.
> It seems we need a patch to change the size of object, e.g. 128K as ZFS.
Originally I planned to use the SHA1 to both name the snapped objects
and regular IO objects. But later I thought the overhead for calculating
the SHA1 for every RW operation would be too costly. So I placed the
those regular IO objects in the 'working directory' in the farm.
It is now kind of relatively easy to add this feature back since we have
got all low level mechanisms of sha1 operation ready. Maybe we could
offer at least one more option to user.
Further more, maybe the whole farm can be implemented as KV store with
data de-duplicated, and the sheep gateway simply talks to object cache
(maybe other tailored cache) for regular IO to speed up the IO performance.
More information about the sheepdog