[Gluster-devel] Improving real world performance by moving files closer to their target workloads

[Gordan Bobic]

Luke McGregor wrote:
> Would there be a way of doing this without broadcasting the request
> for file?

No immediately obvious way comes to mind.

> how does gluster currently handle the file metadata at a low
> level?

I'm not 100% sure, but I think the file's location is dictated by the 
hash of the file's path/name.

> Is there a central or distributed metadata cache?

No.

> Im sort of
> assuming there is some kind of full set of metadata which applies to
> the whole system but from what people have been saying this is not the
> case. How dows gluster work in this regard?

Metadata (e.g. file version for AFR) is attached to the file itself, and 
is stored in xattrs. The only exception I can think of is the striping 
translator, as the file is chunked across multiple nodes, but I'm not 
sure how that works.

Gordan