robert_difalco wrote:

>
> For us, even with forced writes off they seem to be IO bound.
> Basically, each thread handles a batch of updates and inserts that end
> with a commit. With forced writes on, this seems to be faster if
> everything is forced into a single thread (i.e. syncrhonous commits).
> With forced writes off, it seems the best performing number of threads
> is around 4-5. Moving beyond 5 in the thread pool and performance
> begins to degrade again on IO waits.

That makes some sense, since forced writes causes the database to
determine the order of page writes - without forced writes the O/S and
disk system choose the order.

>

For the rest of the issues, I guess I should have asked more questions
first. There are different trade-offs on performance depending on what
else is going on at the same time.

Regards,

Ann