read/write accurate i/o accounting

Yesterday the Linux kernel (2.6.20-pre) just added significant 
improvements to the way read/write i/o accounting
can be measured for a particular task.   This could be particularly 
useful for measurements of the actual i/o
that samba server causes in certain workloads (or even for better 
measuring the Linux cifs client side too,
the actual bytes read over the network, ie not in cache)

See
   
http://master.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=7c3ab7381e79dfc7db14a67c6f4f3285664e1ec2
for more details.