Feature - The deep in 3-D

A two-dimensional representation of a three-dimensional holograph of water. Image courtesy of Peter Hobson, Brunel University

Grid-powered holography 

In recent years, biologists have used holograms to visualize predator-prey encounters, fragile  marine snows (“snowflakes” of detritus showering down through the water column) and other phenomena at high resolution in 3-D without harming the denizens of the deep.

As camera technology improves, images become bigger and more complex and consequently take longer to process and store. A single charge-coupled-device (CCD) chip captures as many as 100,000,000 pixels.

The result? A single desktop machine often takes 12 hours to replay one 3-D image.

Using the grid, Peter Hobson and Henry Nebrensky of Brunel University, U.K., have found a way to replay holographic images of water samples in 66 percent less time than it takes with conventional, single-desktop machines.

Slice and dice

The researchers captured the information about a particle’s size, shape and position in 3-D from the holographic image. Due to the depth resolution required, they replayed the hologram in 2,200 individual, 2-D slices.

With the grid, the time (horizontal axis) required to process a 3-D image goes down dramatically compared to a reference PC. Image courtesy of Peter Hobson, Brunel University

“Processing each two-dimensional slice is a perfect grid application,” said Hobson. “Replaying each slice is independent of the others, needs access to a large amount of temporary storage and is CPU-intensive.”

The researchers divided each recorded volume into 220 packets of ten slices apiece and sent them to several thousand CPUs spread across the seven member institutes of LondonGrid for processing.

Initially, completed slices came flooding in—ninety percent of the slices from the first hologram were returned within forty minutes, or eighteen times faster than with a desktop machine. The final slices were delayed, which led to a less impressive time reduction overall.

Buoyed by preliminary success, the team re-ran the volume processing task. While the grid still beat the desktop, acheiving a significant improvement in rate for the complete volume proved difficult. So they tried something different. They split the replay of the volume into 22 jobs of 100 closely depth-separated 2-D slices. This led to higher return rates, with all slices returned within three hours on one occasion.

While pleased, Hobson believes they can do better. “This first experiment allowed us to understand the strengths and weaknesses of the grid for digital holography. We think we can easily tweak job sizes and submission systems to increase the return rate and also automate the entire process from start to finish, including analysis of reconstructed slices.”

—Dan Drollette and Anne Heavey, iSGTW, with
Peter Hobson, Brunel University

This work has been submitted as an abstract for the All Hands eScience Meeting to be held in Edinburgh in September 2008.