Telescience, Tomography, and the Grid

PROJECT MANAGER
Martin Hadida, UC San Diego

PACI's Telescience alpha project is harnessing the power of information grids to enable neuroscientists to unravel the complex interactions of molecular and cellular biological structures that underlie thinking and learning and give rise to debilitating diseases of the brain, such as Parkinson's and Alzheimer's. Scientists will be able to control a high-energy electron microscope a continent away, capture images of specimens, compute 3-D structural models of the specimens, and deposit these models into a digital library. The Telescience team is constructing this powerful discovery environment from high-performance computers, unique laboratory instruments, and modern information-retrieval systems across the country linked by high-speed networks.

THE TELESCIENCE GRID

NEXT-GENERATION INFORMATION TECHNOLOGY

THE TELESCIENCE GRID

"Telescience represents a new model for collaboration and investigation," said alpha project leader Mark Ellisman, head of the National Center for Microscopy and Imaging Research (NCMIR) at UC San Diego. "Investigators will collaborate over distance, and the grid environment can provide remote, near-instantaneous access to scarce or unique resources, such as the high-energy electron microscopes we use. But it also stimulates our thinking about new approaches to surmount current limitations in information technology."

The current NCMIR telemicroscopy system enables investigators at remote sites to interactively steer the 400,000-volt electron microscope at UC San Diego, acquire images of biological specimens, and perform most of the actions of an operator at the instrument's main console.

The Telescience tool suite is applying grid technologies to link the microscope to distributed parallel computers for performing tomographic reconstruction, a technique that uses a set of image projections at different angles to derive 3-D models of specimens. "We plan to integrate the telemicroscopy and Globus tomography systems, so data can stream from the microscope to the tomography pipeline in real time," said Martin Hadida, Telescience project manager.

NPACI scientists from NCMIR, UC San Diego's Parallel Computation Lab (PCL), and the University of Southern California's Information Sciences Institute have developed a parallel implementation of a tomographic reconstruction algorithm that uses Globus to distribute the task across heterogeneous compute resources. Running on workstations across the UCSD campus and on NPACI's Blue Horizon, the distributed application runs more than 10-times faster in processing tilt series data for neuroscientists at NCMIR and most importantly delivers cycles from the mix of machines to tomography codes "on demand." This achievement will now allow direct coupling of the data acquisition to the computational refinement--returning results in minutes, while specimens are still under investigation in the electron microscope.

A new Web interface helps launch each reconstruction job, and another tool illustrates job progress in real time. Reconstruction and rendering tasks run on Globus, using NPACI's AppLeS and Network Weather Service for dynamic discovery and allocation of resources. The group also is investigating scheduling strategies under Globus for using more sophisticated tomographic reconstruction algorithms, such as "blobs" and maximum entropy.

NEXT-GENERATION INFORMATION TECHNOLOGY

"We anticipate great rewards from current efforts to build distributed cluster computing resources between UC San Diego and SDSC," Hadida said. "Three clusters are being set up on campus--one each at SDSC and the two new Keck Satellite Centers for Research in Biological Structure and Function housed in the Division of Natural Sciences and the School of Medicine. These clusters will provide computational power for our most demanding tomography work, tightly linking computational resources to the data acquisition at the microscope and to distributed databases, leveraging SDSC Storage Resource Broker (SRB) technology and mass storage systems."

Telescience tools are being integrated into a multi-scale database, being developed by NCMIR and SDSC, that will provide long-term data storage as well as organize and provide access to previously recorded data. Technologies from several NPACI DICE projects, including the SDSC SRB, will assist in managing and presenting the information.

The functional requirements are also driving an investigation into next-generation Internet technologies, such as transport protocols for constant-bit-rate traffic, the newest Internet Protocol, and Quality of Service protocols. The Telescience team is working with network engineers to define the needs of telemicroscopy applications to ensure proper operation and efficient use of the network.

Hadida and other Telescience researchers will demonstrate the progress of the Telescience project this July at INET 2000, the Tenth Annual Conference of the Internet Society, in Yokohama, Japan. Their presentation will demonstrate real-time remote control of NCMIR's electron microscope in San Diego from the conference in Yokohama.

"In the demonstration at INET 2000, our software will be running over the next-generation Internet Protocol, IPv6, end-to-end," Hadida said. "In many ways, IPv6 is better than the current protocol, but it hasn't been widely deployed yet in the United States. In April, we successfully tested NCMIR's new IPv6 Web server. This is a major step in preparing for our demo, which will be one of the first uses of IPv6 for a major scientific application." --MG