[ENBD] Ammended setup: -> what would happen if...

[Peter T. Breuer]

"A month of sundays ago Lou Langholtz wrote:"
> The 2.4.20+ md driver at least does the resync automatically in kernel 
> space if I read the2.4.20  code correctly. So no userspace tool needed 

That's correct. Raid has always been that way.

> disconnect but you don't say you'll have multiple writers to the served 
> device... so this is confusing. W.r.t. Coda, well enbd, nbd isn't 
> designed for local block at a time caching. Peter has done an impressive 

If he wants writes to continue locally, followed by a "catchup", he
needs a raid1 setup with one local and one remote (i.e. nbd) component.
This is standard.

> job however at optimizing resyncs when using a client side RAID1 but the 
> concept is still of mirroring two block devices as opposed to a client 
> side local partial cache of the original data ala coda/afs/dfs. There's 
> room to do this kind of stuff though possibly.

It's possible. I'm not sure what the advantage over raid1 would be.

> Have you also looked into GFS (Global File System)? Don't know a good 
> URL for GFS off hand but google will give you some good ones. Maybe it 
> will give you some more ideas??

This may allow him to write on the server and the client both
- something I've been meaning to try for a long time.

Peter