is cache-only-prefixes an nnpfs limitation?

[Tomas Olsson]

Adam Megacz <megacz at cs.berkeley.edu> writes:
> nisse at lysator.liu.se (Niels Möller) writes:
> > One possibly naive idea:
> > ...
> 
> Yeah, I was imagining something similar.  Would this work?
> 
I think so, it's a good idea.  It certainly does help arlad's case a lot.
One would still need to keep track of installed blocks in nnpfs, but the
LRU and the block-to-nnpfs_node mapping could be dropped.  The big question
is if it's worth relaying all accesses to arlad.

/t