On Wed, 2007-12-19 at 15:19 -0800, Dann Corbit wrote:
> The algorithm that I am suggesting will take exactly one pass to merge
> all of the files.
>
>From tuplesort.c:
"In the current code we determine the number of tapes M on the basis of
workMem: we want workMem/M to be large enough that we read a fair amount
of data each time we preread from a tape, so as to maintain the locality
of access described above. Nonetheless, with large workMem we can have
many tapes."
It seems like you are just choosing M to be equal to the number of
initial runs, whereas the current code takes into account the cost of
having workMem/M too small.
We do want to increase the number of runs that can be merged at once;
that's what dynamic run handling and forecasting are all about. But we
want to avoid unnecessary seeking, also.
Regards,Jeff Davis