Query optimization for parallel execution is an open problem. However,
Hong and Stonebraker showed that in the context of XPRS the two-phase
hypothesis seems to be valid. In this paper, we clarify the domain of
validity of this hypothesis which states that ''the best parallel pla
n is a parallelization of the best sequential plan''. To do this, we f
irst clarify the differences between sequential and execution plans an
d make precise the decisions to be taken by a parallel optimizer. Our
analysis suggests that the two-phase hypothesis is usually not valid i
n DM. In SM it is true as long as disks are not involved. Finally, we
used our parallel optimizer to make several experiments which confirme
d some of our conclusions.