Yeah I wrote a script to sum up all of the waits and delays for full and
empty buffers in the daily bptm logs.
Each delay is either 20 miliseconds on average for an empty buffer delay,
and 30 miliseconds on average for the full buffer delays (per veritas).  So
if in the log you see waited for empty buffers 100 times, delayed 1000
times, then you know that there was about 900 * 30 miliseconds extra time
spent just delaying.  You can't really help the waits too much, so an ideal
ratio between delay to wait is 1:1, but in this case its 10:1, meaning you
are spending a great deal of time in delays, and you need to look at doing
some things like changing multiplexing, or fragment size, etc.

So write a script to awk out all of the waits for empty and fulls, and the
also the delays, some those up and compare to each other.  In reality, if
the ratios is like 3:1 or lower for a large backup system, that's probably
pretty good, but if it is like 8 or 9 to 1, look into it...

Taylor