Here at my university, I recently started selling disk space to users
from a server with 4.5TB of space. They purchase space and I make
them their own volume, typically with compression on and it''s then
exported via NFS to their servers/workstations. So far this has gone
quite well (with zil_disable and a tuned up nfsd of course)
Anyhow, the frustration exhibited by a new customer of mine made me
think of a new RFE possibility. This customer purchased some space
and began moving his data (2TB''s worth) over to it from his ailing
RAID array. He became frantic at one point and said that the transfer
was taking too long.
What he was doing was judging the speed at which the move was going
by doing a ''df'' on his NFS client and comparing that to the
existing
partition which holds his data. What he didn''t realize was that the
transfer seemed slower because his data on the ZFS-backed NFS server
was being compressed by a 2:1 ratio... so, for example, although the
df on his NFS client reported 250G used, in reality approximately
500G had been transfered and then compressed on ZFS.
This was explained to him and that averted his fury for the time
being... but it got me thinking about how things such as the current
compression ratio for a volume could be indicated over a otherwise
ZFS-agnostic NFS export. The .zfs snapdir came to mind. Perhaps ZFS
could maintain a "special" file under there, called compressratio for
example, and a remote client could cat it or whatever to be aware of
how volume compression factors into their space usage.
Any thoughts? A quick b.o.o search did bring up and existing RFE
along these lines, so I thought I''d mention that here.
/dale