Private Cloud 19
HPC-as-a-Service: Clouds of Speed
October 23, 2013

In a recent Windows IT Pro webinar, two industry experts highlighted one of the most exciting new developments in the realm of cloud computing: the rise of High Performance Computing-as-a-Service (HPCaaS). In part one of this webinar, Mel Beckman, senior technical director at Penton, provided a brief overview of what HPC and HPCaaS are, why the latter represents such a significant development for organizations in numerous industries, and the key role that the transition from cables to fabrics has played in this progression.

HPC and HPCaaS defined
Beckman defined HPC as any complex system that delivers far more computational power than would be possible with a single compute resource. To really qualify, HPC must be hundreds, if not thousands, of times faster and more powerful than an individual computing system. An HPC essentially aggregates a variety of compute resources into a unified system, which allows users to experience the multiplied power of individual processors. This is a process known as parallelism, and it is critical for overcoming the limits of individual computing machines.

As Beckman explained, HPC is closely related to the concept of the supercomputer, which was effectively a precursor to the HPC. Supercomputers feature server density in the double digits. They offered - and continue to offer - a much higher level of computing power than individual systems, but lack flexibility and are extremely costly, limiting their applicability to only the most advanced, large-scale organizations. HPCs link together hundreds of servers, offering a similar level of performance at a much lower cost. They are frequently used for scientific research, engineering, climate simulation, genomics and other traditionally data-heavy endeavors.

HPCaaS represents the next step in this progression. HPCaaS is, as the name suggests, offered as a cloud-based service. This means businesses and other organizations can utilize as much computing power as they need, and easily scale up or down as those requirements change. Additionally, with HPCaaS, firms can make this computer performance available to any department. and monitor usage.

Beckman noted that hyperscale HPCaaS offers the computing power of thousands of servers. Critically, its power consumption is significantly improved relative to HPC or supercomputers. This is important because CPU chip max speeds are reaching their peak performance, due to the natural limits imposed by the laws of thermodynamics. Massive parallelism is the only way of overcoming this barrier.

Among the most notable applications of HPCaaS, according to Beckman, are big data, mobile information, social networking, and healthcare.

Leveraging HPCaaS
To leverage HPCaaS, firms need to both understand the technology and use the right tools and strategies. According to Beckman, one of the first and most widely-used hyperscale systems on the commercial market is HP's Moonshot.

It is also worth noting the role played by fabrics. Beckman explained that one of the challenges inherent to HPCaaS is the East-West Node Locality problem, which concerns communication of compute nodes. With the right cabling, though, organizations can offer a "torus" with embedded switches. This makes the hyperscale HPCaaS hyperdense and hyperfast, maximizing value and utility for users.