The partitioning requirements for adaptive applications depend on the current state of the application and the computing environment. Rather than the absolute goodness of a particular partitioning technique, it is proposed to base the characterization of partitioning behavior on the PAC tuple. The goal is to capture the overall runtime behavior of the partitioner. The proposed metric for characterizing the quality of a PAC consists of the following five components:

• Communication requirement : This component is a measure of how much the processors have to communicate with each other until the next repartition. Communication can either be inter or intra grid.

• Load imbalance : This component measures the load imbalance created by a partitioner. It occurs when any processor has more than average load and can cause performance to quickly deteriorate.

• Amount of data migration : This is a measure of the ability of the partitioner to consider the existing distribution. This is particularly important for SAMR applications as they require re-partitioning at regular intervals.

• Partitioning time : This component tells how fast the partitioning algorithm computes the new partitioning, given the current partitioning.

• Partitioning induced overhead : This component attempts to capture the quality of the partition generated by the partitioner, using the number of sub-grids and their aspect ratio. Many small grids and grids with bad aspect ratios give rise to high overheads.