> To open up the implementation topic:
>
> My current patch starts a new sequential scan on a given relation at
the
> page of an already-running scan.
I think it should start earlier to make use of the already cached part
of the table.
It is hard to determine, or guess how much is still in cache though.
As a start you could maybe use some (small to be safe) fraction of
effective_cache_size
or shared_buffers.
Andreas