[Linux-cachefs] Limiting case of a big cache

[Dale Pontius]

Short version:  What happens with FS-Cache/cachefilesd as the cache 
grows to the size of the NFS filesystem it's caching?

Where I work we've used a network filesystem to achieve Single System 
Image for years (actually decades) now.  It's been nice, but in the past 
few years it seems that network speeds haven't kept up.  Add to this the 
fact that disk space is pretty darned cheap, and it makes me wonder 
about a slightly different operating regime for a network filesystem.  
We're not using NFS/FS-Cache at work, but I do for my home cluster, 
prompting this question.  I suspect it's a generally applicable 
question, as well.

I'd like to just have enough disk space on my user-facing system(s) to 
hold all of their data.  In this situation, the network filesystem 
becomes a real-time backup, a way to propagate data to different 
systems, a way to keep those systems in sync, and since I mentioned 
backup, a central point to run some sort of "time-machine"-like backup 
of old files.

Are you aware of limitations with the existing FS-Cache/cachefilesd work 
in this type of very-large-cache situation?  Do writes to a cached 
filesystem write to the cache as well as to the NFS server?  Does the 
write operation release when the NFS client has the data, or after the 
NFS server tells the client that it has it?  Does the validation time 
for cached data start to dwarf any cache improvements as cache size 
increases?

On the side, what's the status of the in-kernel AFS?  Last I knew it was 
don't-use, just use OpenAFS, but I thought I heard about GSoC work being 
done with the in-kernel module.

Thanks,
Dale Pontius