High-Performance Computing (HPC) is often considered a boutique capability, or edge case, within the IT community. Frequently associated with in-depth scientific research, HPC is typically seen as the domain of NASA astrophysicists, Department of Energy nuclear engineers, or computational biologists deciphering the genetic basis of disease. However, things are changing as we move squarely into the era of big data and data-driven government; HPC is increasingly being mainstreamed as a day-to-day tool for federal workers.
Over the past 5 years the use cases for HPC have dramatically expanded while the underlying systems have migrated to much more cost effective commodity implementations. Additionally, HPC systems are more and more being leveraged for multi-purpose applications, such as big data analytics, machine learning and remote visualization. Most importantly, HPC workloads are increasingly diverse with a smaller percentage requiring pure performance and many where parallel processing can outperform the most advanced interconnects.
These realities have resulted in different configurations ranging from traditional InfiniBand based raw performance to on-premises virtualized HPC to massive on-demand parallel processing in the cloud. Each being targeted at satisfying a particular workload group. CSRA is employing a converged model that integrates all three with a single user interface so that each can scale independently and resulting in the highest level of system utilization. Utilization equals productivity and independently scalable tiers translate to efficiency and increased agility over time as requirements evolve. The single user interface significantly reduces the learning curve as the system, vice the user, traverses each environment.
The petascale performance of today’s High-Performance Computing systems is highly impressive. With the Presidential National Strategic Computing Initiative (NSCI) Executive Order announced in July 2015, we have embarked on the development of systems that are thousands of times more powerful and capable with increased collaboration across government agencies to drive efficiency and sharing of technology and expanded uses. As these next-generation HPC systems are realized, they’ll be put into the hands of a greatly expanded user base through intuitive user interfaces. Government agencies will then be in a position to take on previously intractable computational challenges and actualize the potential of precision medicine; advanced materials engineering; enhanced climate and geographical modeling and other emerging big data, life science, and engineering challenges. By effectively democratizing this immense computing power, agencies will be able to more effectively and efficiently leverage increasingly complex data sets that can be applied to mission-critical problems.
Nowhere was this generational change in HPC more evident than at the recent SC16 International Conference in Salt Lake City. I was fortunate to attend as part of the CSRA team showcasing our next-generation HPC service model – a three-tiered architecture that combines traditional HPC with cloud computing for on-demand scalability that also enables seamless transition between private and public clouds as customer needs evolve. Our three-tiered approach provides CSRA’s customers with cost savings, customizable and flexible workloads, and optimized performance. The feedback we received at SC16 about what sets our HPC solution apart heavily focused on the unified user interface, because it simplifies working with divergent environments, thus minimizing any learning curve. However, many also reflected it offered an ideal path to the cloud; an incremental evolution as they discovered performance and data transfer implications, one workload at a time. Complexity can often be an inhibiting factor for adopting new technology and our single user interface puts the complexity of multiple system environments behind the curtain.
Over the last 25 years CSRA has worked with many federal agencies, including NASA, the National Institutes of Health (NIH), the Centers for Disease Control (CDC), and the Department of Defense, and partners with leading technology vendors to deliver best-in-class next-gen HPC solutions. With approximately 20 petaflops of compute capacity under management by CSRA and 100 petabytes of on-line storage we have seen a rapid expansion in the past few years. With the explosion of data and the cost reductions of systems HPC use cases that were once out of reach can now be justified.
One striking example is the work we do across the National Institutes of Health to accelerate bioinformatics research through expanded HPC systems and services. In 2001, the initial cost of sequencing the human genome was approximately $100,000,000. Advances in sequencing technology and drastic reductions in cost per gigaflop of processing have driven that cost today to $1,000 per genome. This precipitous drop in sequencing cost has far outpaced Moore’s law, and initiatives like the NCSI will be critical to translating raw sequencing data into actionable medical interventions and enabling the potential of personalized medicine for not only NIH but many other agencies.
In the end, the reason that next-generation HPC is important for the federal government is that it accelerates the processing of enormous amounts of data and, most importantly, shifts the focus from analysis to application. Armed with data sets and the ability to leverage machine or deep-learning methods in the future, agencies can now focus on mission-critical activities and solving our most pressing national and global issues, less constrained by the complexities of the systems supporting their objectives.