BIG DATA
UW Professors Develop ECE-HPC Supercomputer
- Written by: Tyler O'Neal, Staff Editor
- Category: BIG DATA
Four University of Wyoming professors have developed an Electrical and Computer Engineering High Performance Cluster (ECE-HPC) -- on which they have created supercomputing applications ranging from directing rescue robots through burning buildings to allowing physical therapists to monitor their patients’ recovery in real time and from remote locations.
The ECE-HPC essentially is a platform of powerful computing resources for UW engineering professors and their students to plan, design and develop projects that can have a lasting real-world impact.
“Part of what we’re doing is enabling this cluster process for electrical engineering,” says Jerry Hamann, a professor in UW’s Department of Electrical and Computer Engineering.“The computer cluster allows us to do things we weren’t able to do a few years ago. We can use it for problems that take a lot of computational time to solve.”
“We can take that computer cluster and customize it to do all kinds of jobs that you just can’t do on Mount Moran,” says John McInroy, professor and department head of electrical and computer engineering, in reference to UW’sAdvanced Research Computing Center, which is nicknamed Mount Moran. “We can specialize the cluster so it can program tasks a robot can do.”
The ECE cluster, located in a small corner room on the fifth floor of the Engineering Building, contains 12 GPU nodes with 30 Teraflops of computing capacity. A node is conceptually similar to a desktop computer, while a Teraflop is a measure of a computer’s speed and is equivalent to a trillion floating point operations per second.
Because the cluster uses graphics processing units (GPU), it can handle operations much faster than one using central processing units or CPUs, says Suresh Muknahallipatna, a UW professor of electrical and computer engineering.
For example, a model that demonstrates radio waves striking a car antenna would take 3.5 hours to run using CPUs. With GPUs, the process takes a scant 8 seconds, Muknahallipatna says. A model of DuPont Circle, a historic business district in Washington, D.C., that breaks down areas of strong and weak radio reception and cell phone service, would need 40 hours to run using CPUs. By comparison, it takes only 6 hours to run using GPUs.
“We can do these operations in a matter of minutes, rather than hours and days,” Muknahallipatna says.
And that’s necessary, considering the number of applications in the works for the cluster.
Robots to the rescue
One application involves using small, mobile robots to go into dangerous areas, such as a building fire or a military zone, to assess an area before first responders enter.
“We’re planning how you would use robots to patrol and look for fires, what the robot can see and what obstacles are in the way,” McInroy says. “It’s a planning problem. You end up turning it into a problem for optimization.”
“We want to provide a robot with the ability to have situational awareness,” Hamann adds. “We want it to be intelligent enough to make decisions regarding where it’s safe and not safe to travel.”
Nicholas Gurbhoo, a UW master’s student in electrical and computer engineering, demonstrated a computer program that provides various commands for small, four-wheeled robots. During a hallway demonstration, the robots were able to make the same movements in unison or could be individually programmed for different movements.
“In the past, robots had computers actually on them, which was pretty cumbersome,” Muknahallipatna says. “Now, the robots are lighter and more agile, like puppets. The next thing would be to put video cameras on them.”
For example, robots could be sent into a building to determine good locations for transmitter links, says Robert Kubichek, a UW associate professor of electrical and computer engineering. This would allow soldiers, law enforcement personnel or firefighters to establish a communications network before entering a hazardous area.
Aiding physical recovery
Another cluster application allows a physical therapist to remotely view, from a computer camera, a virtual model of a patient moving his or her arm during recovery exercises at home. Using the application, the physical therapist can observe the patient’s range of motion, as well as velocity of movement, to determine whether the exercises are executed properly and to monitor progress. The therapist can observe the arm in a horizontal position; flexed or curled inward; and extended.
“For a physical therapist, it’s an immediate response. They can provide feedback to a patient immediately,” Hamann says.
Additionally, some patients eventually may receive a little assistance from an industrial research robot named Baxter. Through a computer program created on the ECE cluster, the man-size, black-and-red colored robot can help a recovering patient with his or her exercises by actually holding and moving a patient’s arm.
“We can take Baxter and use it for physical therapy,”McInroy says. “It’s really sci-fi, futuristic. The dream of robotics has always been to have a machine, like a human, that can do all kinds of things.”
Because there are a limited number of physicians in Wyoming, a primarily rural state, McInroy envisions the day when Baxter can go into homes and assist patients with their individual physical therapy.
The robot includes cameras built into its face and hands. The hand cameras can be used to record the the patient’s arm movements. The patient can watch his or her exercise movement on the robot’s face camera, Muknahallipatna says.
“Maybe one day, Baxter will be able to move the elbow, shoulder and knee (of a therapy patient),” McInroy says. “That’s the dream. We’re just starting.”
UW purchased the robot from ReThink Robotics, a company in Boston, Mass. Carnegie Mellon University, MIT and Rensselaer Polytechnic Institute also recently acquired their own Baxter robots, which came on the market in May, McInroy says.
UW purchased the $27,000 robot using federal funds previously secured by U.S. Sen. John Barrasso and the late former U.S. Sen. Craig Thomas. During March 2003, when the U.S. first sent troops into Iraq, Thomas saw the potential to use robots for dangerous jobs, including those in the military, and sought funding for UW research, McInroy says. After Thomas died in 2007, Barrasso was appointed to the seat by then-Gov. David Freudenthal and continued Thomas’s funding quest.
Improving signals
Another cluster operation focuses on predicting radio propagation around and over buildings, Kubichek says. Radio propagation is the behavior of radio waves when they are transmitted from one point to another.
“We have a 3-D model of multiple buildings, with a heat map that shows where radio strength is strongest,” Kubichek says. “It can trace radio signals bouncing off of buildings.”
This application can help building users or facility managers better understand the best radio reception areas, as well as dead spots, in a particular structure or group of buildings. Kubichek and Harish Muralidhara, a UW doctoral student in electrical and computer engineering, recently demonstrated a 3-D model of DuPont Circle, which has a large traffic circle around a park at its center. The model pinpointed where transmission signals were both strong and weak in the D.C. neighborhood.
“Tall buildings create an ‘urban canyon’ effect,” Kubichek says of the blocks of structures that can interfere with radio reception and cell phone service. “When you drive through a city and use your GPS, you can sometimes get intermittent service because of the buildings. This (model) lets us do an entire test in one fell swoop.”
“This is becoming important with the growing use of drones/UAVs (unmanned aerial vehicle) over U.S. soil for various applications, such as search and rescue, forest fire detection and commercial aerial surveillance, to mention a few,” Muknahallipatna adds.
For example, during the recent Colorado floods, ground-based cell phone towers failed and mobile versions were brought in to re-establish communications after the disaster, he says.
In another cluster application demonstrated on a computer, electromagnetic spectrums or radioactivity around cell phones was measured. Such measurements can help in designing cell phone antennas, as well as determine where an antenna should be placed on vehicles and airplanes to ensure safe and reliable operation of communication devices, Muknahallipatna says.
“We’re studying how to minimize radiation,” he says.
Hamann adds the data computation the group conducts in the cluster would be appropriate for use in the 3-D CAVE (Cave Automatic Virtual Environment). The CAVE is the centerpiece of the Shell 3-D Visualization Lab in UW’s Energy Innovation Center. One of the 12 cluster nodes runs the 3-D CAVE.
“I think the end product is not the cluster,” Hamann says. “It’s the scientists that know how to use the high-capability clusters.”
“We think it will give us a leg up for attracting other faculty,” says Muknahallipatna, who uses the cluster in his “Multi-Core Programming Using GPUs” course to teach students how to use parallel computing.
While the ECE cluster is currently being used only by the electrical and computer engineering department, McInroy foresees professors and students from other departments -- such as chemical and mechanical engineering -- eventually using the cluster.