The Hermit Hacker <scrappy@hub.org> writes:
> Now, Mitch's results for v7.0 showed something like:
> 0/12 filesystem blocks in/out
> You intepreted that as 12 reads from the file system ... 'out' I would
> have interpreted as writes to the file system, which made zero sense
> ... do we have our 'in/out's backwards here?
Good point. Writes from a SELECT are certainly possible --- the SELECT
could be writing tuple status-flag updates, if it was the first
transaction to verify commit status of tuples created by a prior
transaction. But that again raises the issue of whether we've got
a fair comparison. The 6.5 test apparently only saw already-marked-
committed tuples ...
regards, tom lane