Thanks for the help from everyone on this. Further investigation with
the suggested statistics and correlating that with some IO graphs
pretty much nailed the problem down to checkpoint IO holding things
up, and tuning the checkpoint segments and completion target (128 and
0.9 seemed to be the best tradeoff for me) pretty much cleared things
up.
All told, the run time of this regeneration pass was less than half of
what it was the last time I ran one, despite involving more total
data. Much appreciated.
--
- David T. Wilson
david.t.wilson@gmail.com