In parallel extender, all datasets that are inputs to a single join must be partitioned and sorted on the same keys prior to the join.
In Parallel Extender 6.0, I believe that partitioners and sorts are automatically inserted into the jobs, so that a novice PX user doesn't have to understand partitoning and sorting. However, my recommendation is to turn off the auto- sort/part- insertion and to understand partitioning and sorting before you build a single PX job.
Resorting your dataset is a very time- and space-consuming activity if you have a large volume of data. Also note: typically there is no reason to do a sort on a non-partitioned dataset, so usually any sort in a PX job has a corresponding partitioner.