Rutgers and Xerox Combine Efforts in High-Performance Computing

Traditionally cloistered in academic institutions and research laboratories, high-performance computing (HPC) is making inroads into corporate data centers. Why? Because running real-time business analytics and harvesting hidden nuggets of information from an avalanche of unstructured, big data demand it.

The catch for enterprises is that they lack expertise in high-performance computing at the same time that their end businesses are demanding IT services that require it. So it was natural for Rutgers University and Xerox to form an active collaboration around high performance computing.

"Many people may not realize this, but Xerox already has a very active business providing solutions for enterprise information needs," says Nathan Gnanasambandam, a senior research analyst at Xerox. "There are many more people like me. We primarily research and develop analytics and algorithms that work with high-performance computing, and we serve outside enterprise clients."

That includes financial companies, which call on Xerox for risk assessment, and retailers, which seek data gathering and analysis about consumers. "In many cases, we are analyzing thousands of transactions a day that come in a variety of forms, from online transactions to phone calls to email to paper," says Gnanasambandam. "The challenge for the enterprises we provide HPC services to is that their data centers are organized around traditional transaction processing, and they need formal training in areas like quantitative modeling in order to provide HPC. An organization like ours has the requisite expertise, and can work with these companies in pilot and larger projects on the problems that they need to solve that also require HPC."

Xerox had its own HPC clustered servers in its data center because its algorithms are proprietary and the company wants to own its own infrastructure--but it's also working with universities like Rutgers to see if some of the tasks can be performed on a larger scale by provisioning additional resources when demand requires it.

Partly because of partnerships with companies like Xerox, Rutgers launched its Rutgers Discovery Informatics Institute (RDI2) in March. Out of 62 scientific computation centers in the United States, only eight have industrial partnerships programs--and Rutgers is one. The Rutgers HPC program is designed to not only develop students into future engineers, but also to use its IBM Blue Gene Supercomputer as a collaborative resource with the private sector.

The development of the Discovery Informatics Institute was aided through a partnership with IBM, which provided equipment and research services, with the understanding that Rutgers will purchase additional hardware and software, along with entering into a three-year maintenance agreement for the equipment. The project was meant to be a first step in an economic development opportunity for Rutgers, IBM and the state of New Jersey, which has been engaged in discussions with Rutgers to provide funding for further expansion and supercomputing capacity. For New Jersey companies, the supercomputer will help them bring products and services to market years sooner and less expensively than otherwise possible--and Xerox has been an immediate beneficiary.

Next: How Xerox and Rutgers Work Together on High-Performance Computing"Although we have a large outsourcing business that includes HPC service and our own HPC servers, the collaborative relationship with Rutgers allows us to scale up our HPC computing resources on demand by provisioning from its cloud," says Gnansambandam. "The partnership also allows us to work three or four years with students at Rutgers on high-performance computing projects. Many of these students are at the graduate level, and we have the chance to see their research and work. It gives us easy access to good talent that we ultimately hire."

Michael J. Pazzani, VP for economic development and professor of computer science at Rutgers, adds, "We wanted to create a high-performance computing resource with expert support for industry in New Jersey and the surrounding region. Part of this effort was to improve our HPC education program where students actually have the opportunity to work on real problems that are faced by companies. The other goal of the program is to actively partner with companies in our region on HPC, including smaller organizations that can't afford HPC on their own but still need it in their data centers."

So what makes high-performance computing tick?

"Rapid parallel processing is what sets HPC apart from other forms of computing," says Manish Parashar, who teaches high-performance computing at Rutgers. "Work is collectively done by multiple processors on a given problem, and an HPC application can process different parts of the data in parallel and then piece these various nuggets of data into a final composite. At the university, we use HPC most often in the social and biological sciences. Being able to develop algorithms that answer unique questions and solve unique problems is now an essential skill set in these disciplines."

Gnansambandam said he sees the same thing with outside enterprise clients. "There is a great deal of learning that is simply ongoing," he said. "We begin the process by sitting down with a client in what we call 'dreaming' sessions. In these sessions, we discuss what needs to be accomplished over a five-year period for the business and what the tough problems are. It is our Innovation Group that actually focuses on enterprise problems that seem to have no answers. From here, this group develops new algorithms and also patents. The algorithms are designed to mine data. Sometimes, we will put together a prototype in the lab to show clients how the algorithm works, and what it can bring toward solving a particular business problem."

The research collaboration between Rutgers and private enterprises like Xerox works just like other research collaborations. "Private enterprises and the state of New Jersey provide funding for research and we dedicate post-graduate students and computing resources for the development of solutions for problems that are difficult for enterprises to solve with their existing IT infrastructures," says Pazzani. "The partnership works because high-performance computing demands a different approach to problem solving that many enterprises aren't familiar with, and we bring students who are educated in these skill sets."