IT News & Events

News about IT at Indiana University and the world

Menu
Improving simulations with advanced software library (DASHMM)

IU Researcher Devises Cutting-Edge Technology for Scientific Research with Big Red II

Science has in many cases moved from the physical realm - test tubes and bunsen burners, into the digital. All fields of science now use computing as part of their research, and many of the experiments that are conducted for critical purposes - from simulating weather systems to DNA sequencing - require immense computational power. IU researcher Bo Zhang, the Center for Research in Extreme Scale Technologies (CREST), and his collaborators Benzhuo Lu (Chinese Academy of Sciences), Jingfang Huang (University of North Carolina Chapel Hill), Xiaolin Cheng (Ohio State University), and Jackson DeBuhr are studying ways to make computers more useful and efficient for purposes like these.

Bo-Zhang.jpg

Bo Zhang, IU

Specifically, he has three goals to achieve. The first one is to develop novel numerical algorithms that are able to both reduce arithmetic complexity and communication cost at the same time. To do this, Zhang and his collaborators have created the Dynamic Adaptive System for Hierarchical Multipole Methods (DASHMM) software library. Developed under a NSF-S12 grant, this library is based on an algorithm called the multipole method, recognized as one of the top ten algorithms of the century. It has applications in a variety of fields within computer science, chemistry, physics and many others. DASHMM also contributes to the scientific community by exposing “the characteristics/challenges of the novel algorithms developed by the mathematics community to the computer science community,” says Zhang.

The application of DASHMM plays into Zhang’s second goal: to develop/release open-source scalable packages for large-scale scientific simulations. These packages will be available to any researchers who want to use them, and can be applicable to all kinds of scientific simulations. An example of one of these packages is the AFMPB solver, which Zhang and his collaborators created for computing electrostatic properties and solvation energies of biomolecular systems. With the power of the DASHMM library, they were able to reduce the time to solve of an analysis of the dengue virus from 10 hours to 30 seconds.

1k4r3d.png

Simulation of the dengue virus, aided by BRII


Zhang’s third goal is to develop an asynchronous multi-tasking runtime system to increase the user productivity and performance portability of end-science applications on high-performance computing clusters. Special software like Zhang’s is required to boost the performance of computing clusters, groups of computers that work together as a single system.

To test DASHMM, Zhang performed frequent evaluations and measurements on IU’s supercomputer Big Red II. Big Red II allows him to “deal with the challenges of data placement and migration, priority scheduling for highly irregular and dynamic applications”. These issues are still on-going research questions, and hopefully, with the help of IU supercomputers, Zhang would be able to solve it in the near future.