"Alexander Todorov" <alexx.todorov@gmail.com> writes:
> On 7/1/07, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> As long as shared_buffers is high enough, there doesn't seem to be much
>> point in worrying about this; the incremental performance gain will be
>> minimal since everything will be in RAM anyway.
> Yes it will be but this does not mean there will be no disk i/o
> operations. Database contents still have to be backed up on disk
> (unless there is a mechanism of delayed wrtite to disk which I am not
> aware of).
It's called a checkpoint.
Assuming that you would actually like your changes to get saved
someplace, I doubt you are going to be able to improve efficiency
by replacing the existing write mechanisms by some ad-hoc
application-level backup procedure. That's why I asked if you
thought losing data at crash was a feature, as opposed to a severe
demerit that you put up with in the hope of gaining some performance
--- because unless that's what you think, it's probably not a real
useful path to pursue.
regards, tom lane