No subject

transaction (t_unit_res - t_curr_res), whether the grant head/write
head change apprpropriately, whether there are queued waiters, where
the tail is, where the last iclog was written to, etc. By looking at
the various values, the threshold problem might become obvious.  I
tend to write quick throw-away awk scripts to extract the necessary
fields into a plottable formats and run those through something like
gnuplot or kst to see if there is any visible artifacts that might
be the cause of the problem...

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com