Platform Computing

Before Mississauga, Ont.-based Atomic Energy Canada Ltd. (AECL) switched to a high-performance computing (HPC) technology, the nuclear products and services company relied on costly symmetric multi-processing machines to perform product modeling and simulations.

“We’re modeling nuclear reactions down to the neutron. That’s why something like an HPC environment is very important to us,” said AECL’s manager of infrastructure services, Simon Galton.

Besides reaping the horsepower of HPC in the engineering support of nuclear products, Galton said an environment composed of individual machines allows for scalability and easier replacement of units as technology improves.

AECL made the shift in 2004 at about the time Galton said he observed HPC “had hit a certain stage of maturity” — an essential requirement, considering the nuclear product business is no arena to be using bleeding edge technology.

However, Galton had also been witnessing a better understanding of what HPC technology could do for the enterprise beyond the traditional academic and research sphere.

Universities and research institutes have employed HPC environments for a long time, but the technology has only made its foray into the enterprise in the past couple of years, said William Terrill, associate senior research analyst with London, Ont.-based Info-Tech Research Group.

Adoption was hindered by the fact that enterprise-specific applications for the technology were few, and the development of HPC code was largely proprietary, said Terrill.

Galton agreed that using an HPC environment to perform parallel tasks among many PCs certainly presents challenges, namely the need for tools and code. “The software has to support that, it’s not automatic.”

But HPC adoption in the enterprise today is facilitated by the availability of that code; a message passing interface (MPI) standard for individual computers to communicate; the power of multi-core processors; and an increasing number of software vendors offering tools for computing cluster management, said Terrill.

Al Lopez, vice-president of strategic planning and corporate development with Markham, Ont.-based HPC infrastructure software provider Platform Computing Inc., said although computing clusters are increasingly becoming affordable and accessible, some IT departments may not necessarily have the expertise to manage an HPC environment, let alone build one.

The challenges of integration and management are in part the drivers behind the company’s recent acquisition of the Scali Manage line of business from Marlborough, Mass.-based Scali Inc., which offered an HPC cluster management and monitoring system.

“Having something that’s integrated and works together certainly simplifies the task of putting one of these systems together, and work for you, and then being able to manage it afterwards,” said Lopez.

If the HPC infrastructure goes down, then they have a 1,000 design engineers doing nothing. Nick Werstiuk,>TextHPC may have emerged from academia, but has heavy enterprise use in such areas as semiconductor and automotive design, and in the financial industry to run data-intensive analytics for traders, he said.

In fact, Lopez doesn’t think HPC’s academic roots create a perception that hinders enterprise adoption, evidenced by the sheer number of enterprise uses substantiating its applicability and success.

Actually, there’s been a shift where HPC in the enterprise has become “core to how they do business,” said the company’s senior strategic analyst of corporate development, Nick Werstiuk.

The expansion of the oil and gas industry, he added, has certainly leveraged HPC as a cost-cutting technology.

More mission-critical

Although the applications may differ from those observed in research, said Werstiuk, the technology is by no means diluted for the enterprise. If anything, he said, the application is more mission critical for enterprise customers compared to research’s more experimental approach to technology.

“If the HPC infrastructure goes down, then they have a 1,000 design engineers doing nothing. So we have extra expectation from those customers around the robustness and the reliability of our solutions,” he said.

Terrill agreed that enterprises are not using diluted HPC technology, and in fact, “the technology has caught up with the need.” The infrastructure remains the same regardless of the application, he said, requiring discrete threads between components and brought together by a coordination mechanism. “And that doesn’t change.”

What might change is the use of Ethernet in lieu of an exotic interconnect like InfiniBand, working at a lesser transmission rate, he said. The use of HPC in the enterprise really whittles down to whether a task can be broken into chunks and run independently with little or no communication before being integrated at the end, said Terrill.

Salt Lake City, Utah-based provider of Linux-based HPC offerings Linux Networx Inc. (LNXI), late last year launched an online resource to help businesses and independent software (ISVs) select the right components — processor, interconnect fabric, storage, run codes, and ultimately build an HPC environment.

The trend towards HPC in the enterprise is partly driven by the increasing commoditization of underlying hardware from vendors like AMD and Intel, said the company’s vice-president of technical marketing, Eric Pitcher. “It’s essentially moving down scale of the technology from the high end into the mid range. So that makes it much more affordable for all organizations,” he said.

LNXI’s vice-president of marketing, Kim Wellman, said the Solution Center forms part of LNXI’s goal to help IT managers deploy production-ready, performance-tested HPC technology with little knowledge of HPC and Linux.

The enterprise use of HPC will grow, said Terrill, especially with mundane tasks like filtering through masses of e-mails during e-discovery, for instance. “A large enterprise gets hundreds of thousands of e-mails. And maybe you have a lawsuit and you need to find all references to George Wallis. And when you find it, there are just huge amounts of applications that could be applied to this.”

Related Download
3 reasons why Hyperconverged is the cost-efficient, simplified infrastructure for the modern data center Sponsor: Lenovo
3 reasons why Hyperconverged is the cost-efficient, simplified infrastructure for the modern data center
Find out how Hyperconverged systems can help you meet the challenges of the modern IT department. Click here to find out more.
Register Now