RAxML-Cell: Parallel Phylogenetic Tree Inference on the Cell Broadband Engine

Phylogenetic tree reconstruction is one of the grand challenge problems in Bioinformatics. The search for a best-scoring tree with 50 organisms, under a reasonable optimality criterion, creates a topological search space which is as large as the number of atoms in the universe. Computational phylogeny is challenging even for the most powerful supercomputers. It is also an ideal candidate for benchmarking emerging multiprocessor architectures, because it exhibits various levels of fine and coarse-grain parallelism. In this paper, we present the porting, optimization, and evaluation of RAxML on the Cell Broadband Engine. RAxML is a provably efficient, hill climbing algorithm for computing phylogenetic trees based on the Maximum Likelihood (ML) method. The algorithm uses an embarrassingly parallel search method, which also exhibits data-level parallelism and control parallelism in the computation of the likelihood functions. We present the optimization of one of the currently fastest tree search algorithms, on a real Cell blade prototype. We also investigate problems and present solutions pertaining to the optimization of floating point code, control flow, communication, scheduling, and multi-level parallelization on the Cell.