Abstract: An optimizer for a data transformation system. The optimizer optimizes data flow plans that describe how data is to be transformed from the form it has in a data source to the form required in a data destination. A data flow plan is made up of a sequence of transforms, and the optimized data flow plan is equivalent to the original data flow plan but has fewer transforms. One kind of optimization is read/write optimization, in which the data flow plan is modified so that operations of the original data flow plan are performed in the data source or destination. Another is merge optimization, in which a single merge transform specifies the operations specified in a plurality of the transforms of the original data flow plan. The operations specified in the merge transform can further be performed in parallel. The optimizer additionally reorders the transforms in the original data flow plan to increase the amount of optimization.