Current cifs code doubles file copy speed from server to Linux client - due to Jeff Layton's async read patch set
smfrench at gmail.com
Thu Oct 20 11:58:58 MDT 2011
Did a quick check of the current cifs-2.6.git tree (in linux-next,
which I will be requesting to be merged into 3.2-pre). Thanks to Jeff
Layton's async read patch set, as expected the file copy performance
from server to Linux client is dramatically faster.
In a quick test copying a directory containing 144MB of movie trailers
from a typical Windows Vista desktop system (as desktop) using the
Linux cifs kernel client the speed more than doubled. Performance
should be even better to Linux/Samba servers due to Samba's larger
maximum i/o sizes.
When you also consider that Pavel's byte range lock queueing and lock
batching patches should help some workloads a lot, and that his SMB2
code will allow oplock (lease) upgrades - the performance of 3.2 cifs
kernel client should be much better.
More information about the samba-technical