The University of Maryland College Park is awarded a grant to create tools that expand the computing power freely available to all ATOL (Assembling the Tree of Life), and other phylogenetic researchers. The central idea is to combine Grid and GPU (graphics processing unit) computing to take better advantage of a diversity of computing resources, particularly existing desktop processing capacity available through public-computing. The main product will be a Grid system, customized for phylogenetic tree search and accessed through a user-friendly web interface, that links thousands of individual computers world-wide. The likelihood-based tree search program GARLI (Genetic Algorithm for Rapid Likelihood Inference), will serve as the initial model application during development. The new system is intended to allow analyses that far transcend what one can do on a single desktop computer or moderate-sized cluster. For example, it should become possible to carry out extensive bootstrap analyses in little more time than is now required for a single tree search. The project will also exploit a second under-utilized resource inside desktop computers, namely the graphics processing unit (GPU), which can do many scientific calculations faster than the central processing unit (CPU). A free, open-source library (C/C++) for maximum likelihood calculations using parallel CPU/GPU processing will be developed. GARLI versions will be developed that take advantage of major GPU platforms. Subsequent development will extend to other likelihood-intensive methods (e.g., Bayesian analysis). Five participating ATOL projects will help test the new tools, and publish a collaborative evaluation of the performance of the new system across the diverse phylogenetic problems they represent. The project will be a joint effort with Bowie State University, Coppin State College, and the University of Kansas Center for Research, Inc.
Intellectual Merit BEAGLE (Broad-platform evolutionary analysis generalized likelihood evaluator), an open source application programming interface (API) and software library. Has been incorporated into several popular phylogenetic programs (e.g., MrBayes, BEAST, PAUP*, PhyML). The 2012 publication publication describing BEAGLE has been cited 50 times. A new version of BEAGLE work with OpenCl, supports a broader range of computing resources, provides substantially improved performance, and has been optimized to work on Stampede (an XCEDE resource). A manuscript describing this version 2 of the library is in preparation. A scientific analysis gateway, at molecularevolution.org, has been developed and launched to facilitate submission of phylogenetic analyses to the grid computing system. The gateway includes such features as an elegant user interface, machine learning-based job running time estimation sophisticated, efficient resource scheduling, and automatic analysis post-processing. The system has served 747 distinct users, who have submitted 8115 analysis jobs, comprising 3,678,047 individual search replicates. At least 47 papers have been published since 2010 that mention using the system for analyses. A manuscript announcing and describing the gateway is in preparation. Broader Impacts Training: Direct project involvement included 3 PhD students (1 at HBCU partner institution); 4 Master’s students (2 at HBCU partner institutions); 6 undergraduates; 3 staff at HBCU partner institutions. Outreach: 7 meetings to develop open-source library and associated application programming interface and library; 3 invited seminars by PhD students; 3 conference talks, 1 talk at regional meeting; 3 additional seminars (including 1 at non- partner HBCU). Capacity building: Through the project the first distributed computing environments were established at both HBCU partner institutions. Additional distributing environments were established at the University of Maryland and others expanded. Resource sharing: Through the project four institutions shared temporary excess computing capacity resulting in greater analytical capacity for individual studies without additional computers. Public participation: The project has involved 16,1388 public volunteer users comprising 35,992 public desktop computers in 140 different countries. Environmental benefits: Through the project large amounts of energy were saved through grid and public computing. This resulted in less demand for new computers, which would have to be manufactured and shipped. The use of idle desktop computers, both institutional and public, decreased the energy required for cooling high-density computer centers.
