BSA Past-President’s Symposium: Computational Issues and Solutions for the Study of Plant Phylogeny
Stamatakis, Alexandros , Smith, Stephen A. , Berger, Simon A. , Alachiotis, Nikolaos .
Phylogenetic Inference is an Engineering Discipline.
The field of systematics is currently facing an unprecedented data flood which is driven by novel wet lab sequencing techniques. Concurrently, a second revolution is taking place in the field of computer architectures with the introduction of multi- and many-core architectures and the rise of accelerator technologies such as, e.g., the Intel Larrabee, GPUs (Graphics Processing Units/Graphics Cards), or the IBM Cell processor architecture. In addition, the area of supercomputing is increasingly becoming energy-constrained, which means that only a limited amount of computational resources, e.g., in terms of CPU hours and GB of memory, will be available to conduct large-scale phylogenetic analyses.
Within this context we need to start thinking about how to efficiently exploit a limited amount of resources to obtain optimal results, i.e., introduce the notion of trade-off engineering. Moreover, the adaptation of the computational kernel of the Maximum Likelihood (ML) function to emerging parallel architectures also requires the deployment of engineering approaches, e.g., a detailed assessment whether to use single (32 bit) or double precision (64 bit) floating point arithmetics which can significantly influence program performance and resource requirements in various ways.
I will present various examples for trade-off engineering approaches to phylogenetics, including a bootstrap search convergence (bootstopping) criterion, an ML search stopping criterion, and a detailed trade-off study between single and double precision arithmetics for the implementation of the ML function.
I will also briefly address current work on accelerating the ML function by vector instructions which yields speedups of up to 50% on standard multi-core processors and by building dedicated hardware using Field Programmable Gate Arrays (FPGAs).
I will conclude with a brief overview of future work and describe the cyberinfrastructure tools that will be developed for large-scale phylogenetic inference within the framework of the iPTOL project.
Log in to add this item to your schedule
1 - Technical University of Munich, Dept. of Computer Science, Boltzmannstr. 3, Garching b. München, D-85748, Germany
2 - National Evolutionary Synthesis Center, 2024 W. Main Street, Suite A200, Durham, NC, 27705-4667, USA
Presentation Type: Symposium or Colloquium Presentation
Location: Ballroom 2/Cliff Lodge - Level B
Date: Wednesday, July 29th, 2009
Time: 9:30 AM