Profiling disk usage on selected branches of a file system

[Stephen Satchell <list@fluent2.pyramid.net>]

I have a number of servers that are disk-bound, and I'm trying to convince 
the boss to split out the single disk drive to separate disk drives to 
increase I/O throughput.  I've been using VMSTAT to show just how 
disk-bound we are, but I can't get any traction toward using multiple 
drives until I can show a clear, positive, and immediate benefit.

Searching the archives, I didn't see discussion of methods of monitoring 
separate parts of a file system, other than building separate partitions 
and using /proc/partitions to measure what happens there.  This isn't an 
option.  (There is some political inertia here, as well as a heavy dose of 
NIH.)

Any help welcome.

My hypothesis:  we need three separate drives, one for OS/code/swap/logs, 
one for /home (Web pages, scripts, and user data), and one devoted to 
MySQL/PostgreSQL.  The "right" way to do the job would be to have the 
databases on a completely separate computer, but that's out because of 
proprietary-software issues.

You can see the Catch-22.

Put together a single server to try the concept?  "Please, you must be 
joking,   Our servers are working just fine!"

grumble grumble NIH grumble grumble

Satch