Bruce Momjian <pgman@candle.pha.pa.us> writes:
> The only way I could see it being worse than pre-8.0 is that the
> bgwriter is doing fsync of all open files rather than using sync. Other
> than that, I think it should behave the same, or slightly better,
> right?
It's possible that there exist platforms on which this is a loss ---
that is, the OS's handling of fsync is so inefficient that multiple
fsync calls are worse than one sync call even though less I/O is forced.
But I haven't seen any actual evidence of that; and if such platforms
do exist I'm not sure I'd blink anyway. We are not required to optimize
for brain-dead kernels.
regards, tom lane