HPC Clouds – An Opportunity Not Just for the “Missing Middle”
The term “missing middle” was coined by the Council on Competitiveness and refers to small and medium-sized manufacturers that are missing out on the benefits of advanced modeling and simulation technologies. Capital and operational expenses associated with high performance computing (HPC) deployments required to run simulation codes are a barrier of entry for these organizations.
On the other hand, missing middle organizations are part of a global marketplace that is increasingly competitive. Surviving in this climate requires higher levels of engineering productivity. The use of HPC resources can play an important role for helping missing middle manufacturers achieve these higher levels of productivity as optimizations and design improvements in the virtual concept and design stage can be applied and explored at much lower cost than fixes at later stages in the design cycle.
The on-demand availability of HPC resources “in the cloud” with a pay-as-you-go payment model is a great opportunity for small and medium manufacturers that want to start using HPC resources for advanced modeling and simulation:
- There is no cost prohibitive initial capital investment required for hardware.
- There are no ongoing operational expenses for collocation space, power, cooling and system maintenance.
- HPC Clouds allow for the dynamic allocation of resources and provide great flexibility for project driven workloads and simulation models that scale to varying degrees.
At Penguin we have observed that the benefits of our HPC cloud, Penguin Computing on Demand (POD), are being recognized by an increasing number of small and medium manufacturing organizations. At the recent LS-DYNA conference in Dearborn, MI, there was great interest in using POD for structural FEA simulations and a number of customers are already running simulations on POD. At the same event two years ago there was much less interest and we listened to a lot of users that were concerned about the viability of a cloud computing model for HPC.
Some examples of manufacturers that use POD include IMMI, a medium-sized manufacturer of advanced safety systems for vehicles. IMMI relies on POD for complex FEA simulations that cannot be handled by in-house legacy systems. Callaway, a leading manufacturer of golf equipment is using POD for structural analysis and Waltrip Racing, a NASCAR race team who recently made headlines with Clint Bowyer”s win at the Sonoma raceway, is using POD for quickly turning around CFD simulations. While the rate of adoption of HPC clouds is growing there are real and perceived obstacles that are slowing down this growth.
With or without a cloud, one of the biggest hurdles for almost all missing middle manufacturers that are interested in switching to virtual prototyping is the high barrier of entry for creating digital models of their products. The process of transitioning from physical to virtual prototypes requires significant subject matter expertise and takes time.
An issue brought up frequently as an obstacle to adopting HPC clouds is security. This is probably more a reflection of the apprehension of moving to a new computing model than a concern based on hard evidence of security breaches and compromised privacy that have occurred “in the cloud.” The probability for data leaks within a large manufacturing organization or one of their respective suppliers is several orders of magnitude higher than the probability of confidential data leaking from a “hacked” HPC cloud. With the increasing adoption of the HPC cloud computing model more and more customers will trust their cloud providers security just as most of us trust services like Gmail on a daily basis. For industries that require compliance with regulations such as ITAR (defense) or HIPA (health care) HPC cloud providers already offer compliant infrastructures.
A third concern often mentioned in discussions about HPC clouds is the issue of data transfer. This is a real problem for many organizations. While a transfer of data prior to and after every simulation job may not be an option, there may be other options to resolving the data transfer problem – for example an initial disk transfer followed by ongoing synchronization of input data, remote visualizations of simulation results and application specific compression algorithms. Moreover, network bandwidth is becoming increasingly affordable, a trend that will also help address the data transfer issue.
Another factor that will impact how fast HPC clouds will be adopted in the manufacturing space is the availability of application licenses “on-demand.” The software market in the manufacturing space is dominated by commercial ISVs. With the exception of a few ISVs such as CD-adapco and LSTC, the majority of ISVs has not proposed cloud licensing models. As Earl Dodd pointed out in his DMR article Software Vendors Have Many Tracks Leading to Success in the Cloud, “Independent Software Vendors (ISVs) have neglected—at their own peril—to make the software licensing agreement changes that will allow their applications to run in the High Performance Computing (HPC) Cloud.”
While it is understandable that a drastic change of their core business model scares many ISVs, they will have to adapt to the new reality of HPC clouds. With a very compelling value proposition HPC cloud computing is not a fad that will disappear. ISVs that ignore their customers’ need for more flexible licensing models for the cloud risk falling behind their competitors.
These competitors could be other ISVs as well as open source projects evolving and maturing rapidly into viable, production ready products with commercial support. One example that comes to mind is the open source CFD package OpenFoam that has matured to the point where it is heavily used in production by commercial organizations such as auto manufacturer Volkswagen.
Some software vendors are touting pay-per-use models for their applications on HPC infrastructures under their own control. While this approach supports flexible on-demand use it does not meet the needs of customers who use a variety of tools from multiple vendors in their workflow. This strategy also locks users into a single vendor infrastructure and keeps users from taking advantage of an open marketplace for compute cycles.
ISVs will also have to realize that an early transition to a cloud-friendly pay-per-use licensing model will pay off in the long run. If we look at the missing middle statistics: 95% of the 300,000 manufacturers in the US are of small or medium size. 94% of these manufacturers have not yet adopted high-performance digital manufacturing.
HPC cloud computing is a win-win for the missing middle, cloud providers and ISVs. The missing middle will be able to use HPC resources to realize big productivity gains; cloud providers will grow their business; and ISVs will be able to drastically increase their addressable market and revenue stream.