@michaelcahill, I think this one is yours.

The interesting changes are the incremental backup, of course, but also how we're rolling die to figure out when to do the next checkpoint operation, we've tweaked that over time and this change might break that.

One new feature is the ability to set a timer for the runs, I'd suggest our stress runs should include some longer runs, maybe 10-20 minutes? Right now we don't do longer runs by default (the default is still randomly picking numbers of operations), but I could easily change that if you want.