Hierarchical grid computing is a way to gain high compute power at low cost by combining existing computational resources instead of building a new one. It typically has heterogeneous characteristics, such as: (1) Resources have different computational power; and (2) Resources are shared among users; and (3) Resources are usually connected by networks with widely varying performance characteristics. This makes the development or adaptation of parallel applications on hierarchical grids challenging. In this paper, we study three load balancing techniques for hierarchical grids: static load balancing, master-slave and a new technique called “scheduler-worker”. We evaluate the performance of these techniques for computing the alignment of long DNA sequences on a grid.