With a $9.2 million grant from the Intelligence Advanced Research Projects Activity (IARPA), Professor Andrew A. Chien will lead a team of computer scientists from the University of Chicago who will build the UpDown Systema, a new approach that could speed up graph analysis a hundredfold.
Graph analysis is at the heart of some of today’s most exciting computing applications in science and technology. Organizing data into graphs (large networks of people, molecules or locations connected by their interactions and relationships) can unlock powerful insights for e-commerce, scientific discovery, social networks, recommendation engines and research, and fraud or anomaly detection.
However, today’s computing architectures were not designed for graphs and struggle to be efficient and scalable.
The grant from IARPA, the research arm of the US intelligence community, will fund the development of the UpDown system, to speed up graph analysis. The effort will reinvent computing architecture, dramatically increasing the efficiency and scalability of graphics computing. Such scope will be needed to effectively analyze the world’s largest charts from social media, financial transactions, or Internet of Things device networks that contain billions or trillions of vertices and edges.
“Efficient and scalable computation on massive graph structures is the signature computational challenge for the next few decades,” said Chien, William Eckhardt Emeritus Service Professor in the Department of Computer Science and senior computer scientist at Argonne National Laboratory. “The UpDown architecture we’ve invented has new capabilities to both efficiently encode information and intelligently move it around the machine, two essential elements for faster graphics computation.”
Chien will lead a team of UChicago computer science researchers including Henry Hoffmann, Yanjing Li and Michael Maire; The team also includes graphics computing experts from Purdue University and Tactical Computing Labs, who will design and test the UpDown hardware.
Reverse priorities
For decades, high-performance computer architectures have been optimized to perform linear algebra and other dense operations, with improvements focused on single-value transformations (known as floating-point) instead of the complex, intelligent data movement required by sophisticated data structures. When working with large and sparse graphs, the priorities are reversed: computers perform fewer operations, but must intelligently interpret and move much larger amounts of data.
The UpDown architecture thus allows flexible graphical representation and programmable intelligence to move it around the system. The UpDown Accelerator resides between the CPU and memory, allowing an application to create “software-defined hardware” that customizes how data is encoded, interpreted, and moves through a node. Programmers can write software to increase the performance of specific applications or rely on machine learning to optimize the flow of data through the system.
“The UpDown architecture supports programming paradigms that weren’t viable before,” Hoffmann said. “For example, UpDown can transform data representations on the fly where previous work would have constrained programmers to a single compromise representation. We will apply automated techniques based on our previous work on self-awareness to exploit this opportunity, selecting the best representation at each point. The goal is to intelligently extract maximum value from this new architectural capability.
In preliminary evaluations with Argonne National Laboratory scientists and applications, the UpDown team demonstrated a 100x performance increase across multiple graph computing benchmarks. In addition to speeding up chart analysis tasks, these improvements would also make them more energy efficient.
For the next phase of the project, the team will build on the radical design of the UpDown system, explore new applications, and scale the architecture to tens of thousands of nodes.
“UChicago’s selection reflects the institution’s leadership in large-scale systems, computational architecture and machine learning, as well as the collaborative synergy in these areas at UChicago and Argonne,” Chien said.
The UpDown project was funded under the IARPA AGILE (Advanced Graphic Intelligence Logical Computing Environment) program. Other members of the UpDown team include David F. Gleich of Purdue University, and Dave Donofrio and John Leidel of Tactical Computing Laboratories.
—Adapted from an article first published by the Department of Computer Science
#million #grant #UChicago #computer #scientists #improve #graph #analysis