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

[gordan at bobich.net]

On Fri, 16 May 2008, Anand Avati wrote:

>
>       The look-up of file location is done by the hash. The namespace only serves to
>       present a unified view of all the individual merged stores.
> 
> 
> lookup of a file is done by parallely asking all the servers for the filename. once a server
> responds, the location knowledge is cached till the file becomes unreachable on that node.

Oh, I see! So this is already quite similar to the way it would have to be 
done if probabalistic local caching were used. :)

>       I'm not sure that there is a namespace "cache" per se. I think the file open call
>       is just routed according to the hash.
> 
> 
> The current version of unify does not make use of hashes. A filename could reside anywhere.
> Using hash based on filename has issues with renaming and hardlinks

So the unify translator merely redirects the new file creation to a 
(pseudo) random node?

It sounds like what Luke was talking about isn't as different as I 
thought. :) Perhaps it could be done with a caching translator module that 
just copies a file locally if there is space, or space can be made by 
dropping a LRU file?

Gordan