Professor’s software innovations key to rise of high-performance computing and $6.9B merger
The networking technology that fuels the world’s largest supercomputers and data centers spawned one of 2020’s top tech mergers, and Computer Science and Engineering Professor DK Panda helped make it possible.
[Headshot of Professor DK Panda] PandaA world leader in research on high-performance and scalable communication in parallel and high-end computing systems, Panda was recently featured in an NVIDIA blog post detailing how Mellanox Technologies helped establish remote direct memory access technology as the leading concept for fast networks. In April, graphics-chip maker NVIDIA acquired Mellanox for $6.9 billion.
Panda first started working with Mellanox in 2001 after meeting Kevin Deierling, the newly funded company’s vice president of marketing. Panda explained that his lab’s work on message-passing interface (MPI) software was fueling a trend to assemble clusters of low-cost high-performance computing systems.
Standard networks used prior to 2000 were proprietary and did not provide good performance, Panda said. Then a new networking standard called InfiniBand came into the picture. It was designed by a consortium, which had focused primarily on the hardware.
“[The consortium] never knew how to utilize InfiniBand in high-performance computing systems,” he said. “They had some objectives, but nobody knew how to actually use all the features, how to make the systems work. This is where my group started, we were the pioneers in that.”
After receiving hardware from Mellanox, Panda and his students designed a scalable and high-performance communication library—dubbed MVAPICH—that improves processing by connecting traditional supercomputing software with innovative networking technologies and protocols, thus increasing the data flow speed significantly. MVAPICH delivers the best performance, scalability and fault tolerance for high-end computing systems and servers using InfiniBand, Omni-Path, Ethernet/iWARP and RoCE networking technologies.
His team debuted their open source MVAPICH project at the Supercomputing 2002 conference. Keeping the project freely available is important to Panda, as part of the academic mission and to honor the public funding their work has received from the National Science Foundation and Department of Energy.
As of June 2020, more than 774,000 downloads have been logged on the project's site. It powers the majority of the supercomputers across the globe, including the world's 4th fastest, the Sunway TaihuLight at China's National Supercomputing Center. MVAPICH is also distributed as part of many vendors’ software packages.
“Through our software library, we have been enabling the community to deploy larger and larger supercomputing systems,” Panda said.
Today, this software is used by more than 3,100 organizations in 89 countries worldwide to extract the potential of these emerging networking technologies for modern systems.
“DK’s many years of research effort on MVAPICH communication libraries has made tremendous contributions by enabling many large-scale applications to achieve high performance over InfiniBand networks in big clusters and supercomputers,” said Robert Critchfield Professor of Computer Science and Engineering Xiaodong Zhang. “His impact is not only in the field of computer science, but also spreads over other application areas.”
Since 2001, Panda has extended his research into the emerging areas of big data, deep learning and cloud computing. He has designed high-performance, scalable software for the Hadoop framework, a common middleware being used for big data analytics, which is available to the public.
He is also working with providers to enable his MPI libraries in cloud services.
“That is the next wave,” Panda said. “We have enabled our software stacks on Microsoft Azure and Amazon Web Services. It opens up markets for those cloud providers and opportunities for end users.”
Since their original equipment contribution, Mellanox has been funding the MVAPICH project for the past 13 years, Panda said. Other supporters include NVIDIA, Microsoft, and multiple other companies and funding agencies.
In recognition of his global innovations and contributions to high performance computing, Panda received a 2015 University Distinguished Scholar Award, a 2011 College of Engineering Innovator Award and was named a Rock Star of HPC by Inside HPC. He is a Fellow of the Institute of Electrical and Electronics Engineers. Panda joined Ohio State in 1991.
by Candi Clevenger, College of Engineering Communications, email@example.com