Interview with Barbara Hutchings, ANSYS

Barbara Hutchings is Director of the Alliance Program at ANSYS. This programme involves partnerships with significant HPC players including hardware and software vendors and academics. PlanetHPC asked her about HPC activities at ANSYS.

In which application areas do you use HPC?

ANSYS is an ISV, supplying engineering simulation software across a wide variety of industries, for a range of applications. ANSYS simulates everything from gas turbine engines, to automotive aerodynamics, to plastic bottle manufacturing, to wind turbines. Just about any engineered product will benefit from simulation-based product development. These simulations involve Computational Fluid Dynamics, Finite Element Analysis, Computational Electromagnetics and Multiphysics. HPC is important because it enables more detailed and accurate simulations to be accomplished, and more of them, so that engineers can get greatly enhanced insight into the behaviour of a proposed design.

Why do you use HPC?

Our customers use HPC to increase the strategic value of simulation. Simulation is typically less costly than physical prototyping. The other important issue is that it can be accomplished very early in the design process before physical prototypes are available. This is the most effective (lowest cost) time to consider innovative ideas. Making changes later in the design process, after physical prototyping, is much more costly and often the options are then more limited. HPC simply adds to this value proposition, by enabling a more accurate, more detailed, and more complete simulation-based understanding.

What are the cost benefits to your business of using HPC?

The cost benefits of HPC to our customers include the ability to:

• Make the "right" design decisions, early in the process, so that the overall design process is streamlined (shorter) and less costly.
• Get the product performance "right" and thereby avoid warranty costs or product failures that may be very expensive to fix in the field.
• Innovate and succeed in the market (vs. the cost of a product that fails).

Again, this is essentially the same value proposition that simulation offers in general. HPC, however, increases the likelihood of successful outcomes in each of these areas, because it enables high-fidelity understanding that can be impossible to get any other way.

Which HPC systems do you use in your business? What is their capital value? How scalable are your applications?

ANSYS customers typically use distributed memory clusters or SMP servers or both. High-end desktop workstations are also commonly used. ANSYS CFD simulation software will scale to 100s or even 1000s of cores. ANSYS structural analysis software will typically scale to 10s. Of course, HPC systems are also used to enable simultaneous consideration of multiple design ideas, for example several simulations running concurrently, each using 10 or more processor cores.

For your business does the Cloud offer a viable alternative to owning and managing your own systems?

ANSYS customers may benefit by using the “Cloud” to access the hardware cycles they require. There are challenges related to file transfer (large datasets) and data security, but these issues can be overcome.  Cloud solutions that offer access to hardware, for example IaaS (Infrastructure as a Service) providers, could be very valuable especially for SMEs (Small to Medium Enterprises) with little or no expertise (or interest) in managing HPC systems.

An important issue here is licensing, and ANSYS is very flexible in this respect. Our customers buy a software licence and they can run that software wherever they want, including remote “Cloud-based” hardware. We can also offer licences to cover peak-capacity demands. Essentially, we maintain direct contact with our endusers and can design a licensing strategy to support Cloud-based computing.

What are the challenges you see in the development of your HPC capability (e.g. scalability of applications, power consumption, cost of systems)?

As a software developer, the primary challenge for ANSYS is to keep our software current as HPC technology continues its rapid evolution. We work closely with the major technology providers and invest significantly in software development to optimise performance, adapting to the continuous developments in HPC architecture. In addition to raw scaling and performance, we are looking at issues such as ease-of-use, robustness, and data management. And we have to look at how HPC and parallel processing can impact the entire simulation process, from model creation, meshing, solution, through to postprocessing.

Are new languages and programming paradigms needed particularly as we move toward exascale systems?

The predominant HPC technology used by ANSYS customers today is a cluster of 64-bit x86 processors linked via a high-speed interconnect such as  Infiniband or 10gigE. For the most part, our software is based on MPI and OpenMP protocols, for distributed-memory and shared-memory execution. We invest significantly to tune our codes for scaled up execution and have demonstrated scalability out to 1000s of cores for some applications. However, this is a constant software development effort, with new algorithms and tuning required to achieve the next level of scaling required by our customers. As processor core counts increase, for example, we anticipate the need for hybrid algorithms using MPI across clusters and OpenMP across multicore devices.

What are your views on GPGPU computing?

GPGPU technology is a good example of how changes in HPC infrastructure will require on-going software development. Due to hardware architecture issues, primarily the GPGPU memory capacity and access time, it’s not possible to simply apply our existing software algorithms to GPGPU computing. While there are isolated portions of our code that can be relatively easily executed on GPGPUs - and we have demonstrated success in these areas – more widespread support for GPGPU technology will require a significant investment. A key issue is thus to understand the return on investment for GPGPUs vs. traditional processors,  for ANSYS and our customers. This is made more challenging due to the ‘moving target’ nature of the technology.

Describe the HPC systems you would like to have available in 3, 5 and 10 years' time.

We envisage systems with more and more cores, running into to 10,000s of cores. For such systems throughput will be as important as peak performance. That is multiple copies of the same code may run on a system, for example, running Monte-Carlo simulations where variations of input parameters are used to develop a statistical understanding of systems. Of course, scalability will be an important feature of codes which run on such systems

In which HPC research areas would EU-funded programmes benefit your business?

There is clearly a role for government funded research into basic algorithms and technologies that will enable software scaling to extend to the next level, and for programs that demonstrate this scalability on important ‘challenge applications’.  Perhaps the most important industrial need is to encourage adoption of HPC, particularly by SMEs who can take significant advantage of advanced simulation enabled by HPC, but may not be aware of the benefits and may not have the expertise to manage a HPC infrastructure. It is important to demonstrate to them how they can benefit from HPC-based simulation. Intervention by government agencies to encourage and support the wider use of HPC for simulation could have significant economic benefit.   We have done some good work of this sort under EU-funded projects.  One particularly interesting project is the @neurIST (www.aneurist.org) project, which included the demonstration of large-scale HPC to enable simulation of cerebral aneurysms, with the goal of assisting clinical diagnosis. This project addressed interesting IT issues such as data mining, security, and data integrity, as well as ease of use.

Our thanks Ms Hutchings for her interesting and cogent comments.

Comment on this interview