Preparing for the HPC Floodgates to Open

On-demand hardware and pay-per-use licensing are essential in opening new markets for HPC.

Bright Computing’s Cluster Manager gives users the ability to monitor HPC clusters in real time. Image courtesy of Bright Computing.

Frame Virtual desktop infrastructure provider Frame enables teams to rent high-performance hardware for intense computing applications. Image courtesy of Frame.

If Adam O’Hern has something he needs to render urgently, “I can buy a $5,000 machine to do it myself, or I can do it for $3 on Frame,” he says.

O’Hern, founder and head of Design at Mechanical Color LLC, specializes in virtual photography. To create iconic images for glossy fashion and lifestyle magazines, a photographer like Annie Leibovitz would set up the scene in a studio with live models. By contrast, a virtual photographer like Adam would set up a CAD model inside 3D software to produce computer-rendered images that are virtually indistinguishable from real photos. In both cases, skill is not the only barrier to entry. Just like Leibovitz has to invest in expensive camera equipment, a virtual photographer, too, needs to acquire and maintain powerful hardware to render. Except, in O’Hern’s case, he now has the option to rent the hardware he needs from Frame, a VDI (virtual desktop infrastructure) vendor.

When asked about his company’s size, O’Hern chuckles. “Me — that’s my company’s size,” he says. “When I need scale, I hire contractors.” Even though O’Hern is in the Washington D.C. Metro area and his contractors are located elsewhere, they can still share the same machine, after a fashion. The so-called machine is a virtual workstation housed in Frame’s cloud server, accessible 24/7. So long as both he and his contractors aren’t working on the same project at the same time, Mechanical Color doesn’t even need an additional software license. He just needs to share his log-in credentials to the virtual machine with his contractor.

“Five years ago, there was no way I could have operated like I do today,” O’Hern reasons. “It’s because of tools like Frame that I’m able to be a small shop with low overhead. Otherwise, I would need IT people, regular employees and in-house hardware.”

O’Hern and Mechanical Color represent the new blood in engineering. Driven by on-demand computing power and ad-hoc design teams, these small shops compete with the big guys on more equal footing. They were previously an underserved segment, but recent trends indicate hardware and software vendors are finally making the much-needed reforms to cater to them.

Small Shops with Big Jobs

O’Hern counts recognizable household brands among his clients. He and his colleagues may be responsible for the eye-catching advertising and marketing imagery that convinced you to splurge for the latest game console.

“Before I began using Frame, I used AWS (Amazon Web Services), but that meant I had to deal with the IT stuff that goes with it — setting up nodes, setting up remote desktops, specifying IP addresses and so on,” recalls O’Hern.

What O’Hern went through was the classic HPC (high-performance computing) adoption pain. Even getting his HPC from the cloud didn’t completely eliminate it, because a general purpose on-demand cloud vendor like AWS didn’t have a customized product for O’Hern. Frame, on the other hand, offers virtual machines for personal use and teams, and VDI hosting services for larger enterprises.

“With Frame, you can rent the equivalent of a $13,000 workstation for an hour for less than the cost of a Starbucks Latte,” says Justin Boitano, VP of Marketing at Frame. That allows O’Hern to pay for more power when he needs it to speed up his jobs — an option he doesn’t have with a real machine.

“I’m working on an animation project that I would have previously had to render overnight on my own workstation,” O’Hern explains. “But now I just set it up on Frame and keep it running super fast in the cloud. At the same time, I can continue to work on my own desktop, because it’s not tied up with the rendering job. Jobs that used to take weeks to finish in series can now be done in parallel in just hours.”

O’Hern thinks it’s only a matter of time before cloud-hosted HPC becomes the norm. At the present, however, he is acutely aware he must accommodate some clients whose policies prevent him from harvesting the Cloud. “Every client is unique. Some have their own security protocols. That’s understandable,” he says.

IT-Free Products from an IT-Free Vendor

Onshape, founded by a team of veterans from the CAD software business, came online in early 2015. Previously most CAD vendors argued it was impractical for them to deliver — and for the users to use — a full-fledged 3D mechanical modeling program in a browser (like a social media app or a multiplayer online game). But Onshape defied this notion with its flagship product, a parametric CAD program that runs in the browser, sold as a subscription.

Onshape runs the company in a minimum-IT environment. “We use Gmail, Google Drive, Slack, Jira, GitHub — we use Cloud tools for everything. The tools that we use to manage our cloud tools are also in the cloud,” said Dave Corcoran, Onshape’s VP of R&D.

“Our entire business runs on AWS,” says John Rousseau, Onshape’s operations team lead. “For all our collaboration, documentation, development tracking, and product testing, we use cloud-based products.”

At the time Onshape was founded in 2012, on-demand computing services were available, but the idea to run an entire company in the cloud did carry certain risks. “Infrastructure as a service (IaaS) has come a long way in the last five years,” said Corcoran. “What Amazon provided back then and what it provides now are very different. And the cost keeps coming down.”

As a vendor delivering its product from a browser, Onshape has to deal with something desktop CAD vendors don’t have to — keeping the product’s performance consistent during peak times and lull periods. “There are peak moments during the day; there’s a drop at night. We have to scale up and down in response to balance performance and cost for our customers globally,” Rousseau explains.

Those peaks and drops also vary across different time zones around the world. “[The demand for service at] 2 p.m. EST is completely different from 2 p.m. PST,” Rousseau adds. An in-house cluster with finite computing power would have a difficult time coping with the varying online traffic. But with an elastic IT setup on AWS, Onshape adjusts its own backend system in the Cloud to respond to the changing workloads.

Bursting at the Seams

Previously HPC was the exclusive domain of large enterprises and university researchers. To harvest the supercomputing capacity of clusters, a business must have a generous budget to acquire the necessary hardware and a robust IT department to manage it.

Bill Wagner, CEO of Bright Computing, identified the three classic hurdles to HPC:

1. Building the cluster.

2. Managing changes in the cluster.

3. Monitoring the cluster health.

All involve a fair amount of time and human resources. Take the cluster health monitor, for instance. “That is a 24/7 job,” Wagner says. “You have to keep tabs on how the cluster is performing. Are there nodes that are not performing? Have they gone dead? Do they need to be rebooted?”

On-demand HPC vendors and open source HPC management software reduces the cost barrier, but not necessarily the complexity, Wagner cautioned. Though a commercial HPC software vendor, he doesn’t see open source HPC software as a competition. Quite the opposite — he says: “We often find that the clients who have attempted to build and manage a HPC setup using some open source products are our best prospects. Typically, open source HPC is a collection of discrete tools, each solving a piece or providing a specific function. So users are left to cobble these together and manage the interaction among them. They quickly find out how complex it is.”

Bright Computing Bright Computing’s Cluster Manager gives users the ability to monitor HPC clusters in real time. Image courtesy of Bright Computing.

Wagner has been noticing some new trends. One of them, he said, is those who want to augment their existing HPC setup with on-demand products. “We’re seeing a pretty sharp increase in piloting and implementing hybrid environments — on-site HPC with the option for cloud bursting on Amazon or Azure,” he explains. “With these companies, their hardware is finite, but their workload periodically rises significantly. But they do not want to build an HPC setup to meet the maximum workload. That usually leads them to their first foray into the public cloud.”

Bright Computing is also anticipating the rise of a new type of customers — those whose business model is to provide on-demand HPC services to smaller shops and individual users. “We’d like to make it practical for anybody to build a cluster,” says Wagner. “One of the capabilities in our product is to create a cluster as a service.”

The Shift from High Margin to High Volume

Juan Betts, managing director of Front End Analytics, is witnessing — and helping create — the emergence of app-based simulation. Until now, the use of CAE was confined to experts in large aerospace and automotive firms because general-purpose simulation software requires a high degree of skill to use. Betts and his colleagues specialize in developing simulation templates and apps that can easily be used by those with limited knowledge of general CAE products. Deploying simulation apps for a wider user pool usually involves setting up the backend HPC infrastructure to support it.

“Larger companies have their own IT systems. They just want us to build the apps they need. They tend to deploy the apps behind their own firewalls. But SMBs don’t have the hardware or the IT expertise. That’s where we see the push toward the cloud,” says Betts.

One of the hurdles, Betts revealed, is the prevalent CAE licensing method. “About 80% of the CAE space is dominated by automotive and aerospace companies. In some cases, you have a single engineer using more than $100K worth of software. So you’re dealing with business models that cater to a relatively small number. With the SMBs, the number of potential users is much greater, but the per-person cost they’re willing to spend is relatively low. The CAE software vendors have not figured out how to deal with that yet,” he notes.

SMB use of CAE is also different from their larger counterparts in frequency. Whereas large aerospace and automotive manufacturers have sufficient CAE work to justify a permanent, on-site installation, SMBs use CAE only periodically. Betts said, “Some SMBs say, ‘I can’t build and maintain an HPC infrastructure for something I just use about two weeks per year.’ What they need is an on-demand model. They need the capability to cloud burst.”

Betts hopes CAE vendors recognize that, even though the per-user cost SMBs are willing to spend is significantly lower, the total spent by this segment could add up to large revenues due to the larger user pool. He has reasons to be hopeful; he’s beginning to see CAE vendors introducing new pay-per-use licensing options.

A Reformation in CAE Licensing

Wim Slagter, ANSYS’s director of HPC and Cloud Marketing, and his colleagues have been pondering the pay-per-use model for some time. As this article goes to press, they’re getting ready to announce a new option, dubbed ANSYS Elastic Licensing.

“It’s like buying a phone card preloaded with a number of hours of Elastic Units. You can use it to pay for any ANSYS products,” he says. The new option is an addition to the existing annual licensing, short-term licensing and perpetual licensing models.

ANSYS Elastic Licensing does require an initial investment — that is, the purchase of a minimum quantity of Elastic Units. When the user has depleted the units in the account, they can purchase additional units on demand. Slagter said ANSYS Elastic Licenses will initially be offered in conjunction with on-demand HPC services from cloud-hosting partner providers; therefore, businesses that do not want to invest in their own hardware, or simply want to augment their existing hardware with on-demand computing, may turn to ANSYS HPC partners.

At the Threshold

Front End Analytics’ Betts think wider adoption of HPC-driven CAE — the kind that involves automatic exploration of hundreds or thousands of design options — depends on three key elements: on-demand hardware, pay-per-use CAE licensing, and simulation apps. The first already exists in the market. Betts says the CAE industry is slowly moving toward the pay-per-use model. And he and his colleagues are in the thick of the last item: developing simulation apps.

“On-demand storage capacity has found acceptance. I think on-demand computing will become the same,” Mechanical Color’s O’Hern says. “Right now, for most companies, cloud-hosted data storage is just a normal part of business. I think the same thing will happen to serious computing power.”

Bright Computing’s Wagner says: “HPC is not only reaching the lower end of the market, it’s also finding its way into industries that don’t traditional deploy HPC. These new users are discovering that the applications they run are dealing with much more data, and the computation required is much more intense.”

More Info


Share This Article

About the Author

Kenneth Wong's avatar
Kenneth Wong

Kenneth Wong is Digital Engineering’s resident blogger and senior editor. Email him at [email protected] or share your thoughts on this article at digitaleng.news/facebook.

      Follow DE
#14954