FLSmidth transcends on-premises limitations with Azure high-performance computing

When companies around the world need sustainable and cutting-edge engineering services in the cement and mining industries, they turn to FLSmidth, the Danish multinational company that sets the pace for sustainable engineering of solutions, services, and equipment.

Founded in 1882 as a builder of small machinery for local craftspeople, the company became an international engineering concern just eight years later. Based in its original Copenhagen headquarters to this day, FLSmidth has branched out its operations to 60 countries. The FLSmidth engineer of today applies the same ingenuity that characterized the company founders, but the scope of the problems that today’s engineers solve calls for enormous quantities of computing power to support sophisticated modelling. That’s why the company engaged with UberCloud, a Microsoft Partner Network member that provides a cloud simulation platform for engineers, choosing Microsoft Azure high-performance computing (HPC) and supporting Azure technologies to achieve the agility, scalability, and performance it needs to deliver the best possible results.

Challenging itself to deliver sustainable, cutting-edge solutions

FLSmidth is on a mission. It’s out to meet the twin challenges of providing the technologies needed to process the cement and minerals demanded by surging urbanization and economic growth and helping its customers reduce environmental impact. The company’s MissionZero sustainability initiative seeks to propel FLSmidth mining and cement customers toward zero emissions by 2030. FLSmidth developed the initiative as an extension of its commitment to the Science Based Targets pledge, which calls for the company to be carbon-neutral in its own operation, dedicate 30 percent of its spend to suppliers with congruent targets, and reduce customer-associated emissions per revenue by 56 percent. The innovation supporting its goals and deliverables for its customers requires commitment—and sophisticated technology. 

Computer-aided engineering (CAE) software is a boon to engineering teams, who use it to simulate the real-life phenomena that will affect their designs. Those simulations require massive amounts of computing power and powerful engineering workstations. Simulations that are run on-premises can quickly bump up against compute limitations—at the cost of speed and performance. Initially, the company depended on a single central HPC cluster in its Copenhagen datacenter. The distance between the datacenter and remote FLSmidth offices in India and other locations slowed performance for those teams.

Engineers at FLSmidth thrive on challenge. What they didn’t enjoy was waiting for finite access to compute time or dealing with the constraints on performance brought about by the limits of workstations and servers. And engineers at the company’s remote offices chafed at the latency and performance issues that came with the distance from the Copenhagen datacenter, where the on-premises HPC system resided. The need to allocate central processing unit (CPU) resources worsened the problem. “We had a resource management system on our HPC platform that queued simulation jobs,” explains Sam Zakrzewski, Fluid Dynamics Specialist at FLSmidth. “We had issues with scenarios that demand elasticity—if we had a sudden large requirement for a project that depended on a lot of CPUs, our only option was to stop all the other jobs on the CPUs we needed, or we had to compromise on the CPU amount we would use.”

Neither did the on-premises model play well with advances in technology. On-premises users who didn’t have desktop workstation devices with high-performance (and expensive) graphics processing units (GPUs) couldn’t access innovations in engineering applications that relied on GPUs. Servers were nearing end of life. “We were at a sticking point in terms of devices,” recalls Zakrzewski. “We had to decide whether to continue capital expenditure investment in an aging system or opt for the cloud.”

But solving those issues with the software as a service (SaaS) approach initially raised questions at FLSmidth, recalls Zakrzewski. “There was a lot of discussion at first about data safety, legal issues, and compliance with data sovereignty,” he says. It became clear that Azure could address all those concerns. “We could also use the power of Intel processors within Azure and thus not compromise on our performance requirements,” he adds. Weighing ongoing costs for an on-premises solution that was beginning to show its age against the possibilities of newer technology, the company revisited cloud computing in 2018.

Threading digital transformation throughout the company

Having historically teamed with his enterprise IT colleagues to research and assess HPC solutions, Zakrzewski found that cloud technology advances and the FLSmidth innovative drive had sparked the same inspiration for both groups. “We want to transform digitally,” explains Holger Kirketerp, Head of Cloud Center of Excellence at FLSmidth. “And as part of our alignment with the business side of the company, we wanted to support our engineers with a system that offered improved performance and flexibility.”

Kirketerp cites the usual reasons for cloud migration: simplifying infrastructure to reduce both hardware costs and management time, redeploying operations staff to more value-added roles, and boosting collaboration alongside performance. “The possibilities we can access with the cloud are incomparable to what we had with on-premises platforms,” he says. Of course, some applications need to be adjusted for optimal cloud migration, and the IT team also had to contend with business systems and engineering workers spread across the globe. The company decided to re-evaluate its decentralized business applications together with its engineering software and create a unified cloud strategy that would enhance productivity for everyone. The way forward: a proof of concept (POC) based on the engineering team requirements. Choosing the vendor was the all-important first step.

Finding the perfect vendor duo for a specialized need

A long-time Microsoft customer, FLSmidth relied on Azure for data storage for some systems and on Microsoft 365 apps for knowledge worker productivity. Choosing a Microsoft partner that specialized in HPC for engineering applications was the obvious course. “It was a natural next step to work with Microsoft to assess HPC for our engineering simulation needs,” explains Zakrzewski. “We contacted UberCloud and created a POC on its Intel-powered, Azure-based engineering platform.”

From a corporate IT perspective, that decision resonated with Kirketerp. “We’ve tried other cloud providers, but nothing has ever matched Azure capabilities,” he says. “When we briefly switched our POC to another cloud in the middle of our selection period as a possible cost-saving strategy, we found that the level of support wasn’t up to that of Azure. We switched back.”

He and his IT team worked with Zakrzewski as a subject matter expert and the UberCloud team to establish the POC. FLSmidth engineering teams in India, South Africa, and Brazil worked with the Copenhagen team to define requirements. The HPC solution would have to support specialized engineering simulation software, or solvers, notably Siemens Simcenter STAR CCM+, Rocky DEM, and Ansys CFX, Ansys Fluent, and Ansys Mechanical.

The highly successful POC marked the signpost toward full implementation. The project team’s deployment plans then accelerated as a key server approached its end of life. With engineering productivity to safeguard, the blended UberCloud and FLSmidth teams doubled down on their timeline.

Building a better mousetrap—with Azure as a base for the UberCloud solution

The UberCloud team began by helping the FLSmidth engineering teams move their engineering simulation workloads to Azure. Kirketerp’s team from FLSmidth closely collaborated with Zakrzewski and other engineers, plus the UberCloud team. Together, they implemented the UberCloud HPC platform in just two months. “We owe our quick deployment in part to the expertise of the UberCloud team,” says Kirketerp.

The combination of FLSmidth IT staff and engineers with the UberCloud team and Microsoft Global Black Belt technical professionals (highly trained Microsoft experts) proved a highly effective trifecta. “The Microsoft solution architects and Global Black Belt team were key to our success because our platform sits on top of Azure,” says Reha Senturk, Enterprise Solutions Director at UberCloud. “We collaborated with them to design an optimal architecture for the best possible performance. And that architecture relies on a full range of Azure solutions beyond Azure high-performance computing—including storage, archiving, monitoring, and reporting.”

The UberCloud solution uses Azure as the base for the flexible HPC cluster architecture at FLSmidth. Engineering data resides in Azure NetApp Files and Azure Files storage, with visualization performance optimized by Intel processor–based HC-series virtual machines (VMs) for GPU, using NVIDIA V100 GPUs. In fact, since the switch to Azure, FLSmidth engineers realize 10 times faster results from their Rocky DEM solver. “We’ve advanced from HC-44 VMs to the current fourth-generation hardware now in use at FLSmidth,” says Pär Persson Mattsson, Head of Sales for Europe at UberCloud. “UberCloud and Azure have grown in tandem, creating greater value through continually higher performance for our customers.”

Engineering teams like to focus on the challenge of bringing innovative solutions to life rather than on the work of creating and managing HPC clusters. The management simplicity built into Azure CycleCloud makes it easy for FLSmidth engineers to optimize HPC regardless of scale. “Azure CycleCloud is a very user-friendly way of provisioning HPC resources,” says Senturk. “That’s why we combined it with our UberCloud Engineering Simulation Platform. It only takes a couple of quick steps for an engineer to access a cluster. Within minutes, they have new resources available to them—and it reduces overhead on IT to provision resources.”

FLSmidth engineering teams experienced a smooth transition. “We’ve had a very positive experience with the cutover to the new system and the enhanced usability,” says Zakrzewski. “The interface for our new Azure-based system is virtually the same as the one our engineers used on-premises. They simply sign in to remote sessions on the UberCloud containers.”

Democratizing HPC management

HPC solutions require specialized support that usually lies outside of an enterprise IT department. Having fallen into essentially a support role, Zakrzewski was concerned about single-handedly meeting that need for his team. As the person with the deepest experience with HPC systems, he didn’t want to be a bottleneck for his fellow engineers—ultimately about 2,000 people who need to stay productive—and he worried about project delays, should he be unavailable.

The fusion of Azure technology with UberCloud negated that concern. “As soon as we began using Azure and UberCloud for our HPC applications, we had a democratized platform,” says Zakrzewski. “It may sound like I’m quoting taglines, but it’s true that as soon as you move a complex system into an enterprise platform like UberCloud, it becomes a lot easier for IT to understand it. And by the same token, it’s easier for engineers to access it.”

That cloud benefit also translates to breaking down the functionality and licensing silos inherent to on-premises hardware. Free from the restrictions imposed by device-based licensing, FLSmidth engineering teams have greater access to their specialized simulation software, and the ensuing performance gains heighten efficiency. “Our cloud-optimized applications run much better and faster in the cloud,” says Kirketerp. “And they’re more reliable.”

Zakrzewski elaborates from the engineering point of view. “Obviously, now that we’ve moved to Azure, we move data fast because of the massive backbones Microsoft has created,” he says. “We engineers no longer experience latency issues because we have the network availability we need to support our work.”

Facing the future with a flexibility-first platform

FLSmidth regarded the project as an opportunity to optimize its systems for growth, and the cloud offers the freedom it needs to respond to a changing world. “There were so many arguments in favor of moving to Azure,” says Kirketerp. “The engineering teams can work in a more dynamic, flexible way. Meanwhile, from an IT point of view, we are free from managing more hardware. And when FLSmidth acquires new companies, we know that there will be a nearby Azure datacenter, wherever that company is.”

Although his engineering colleagues are famously taciturn, Zakrzewski can attest to their satisfaction with the solution. “Our engineering team likes the versatility that Azure offers,” he says. “We can access different instances as we need to, and our colleagues in India now experience low latency for their visualizations. We’re very happy with the cloud experience.”

The success of the cloud migration sparked interest from business application users in the FLSmidth ecosystem. “We share competencies all over the globe,” says Kirketerp. “Although it wasn’t originally in our scope, we are supporting their migrations. They’re excited about the cloud because they see the advantages of that environment.”

For both engineering and IT teams, the migration met every goal. “The Azure platform powered by Intel architecture has been an excellent match for our needs,” concludes Kirketerp. “The capabilities we have and the results we’re achieving show us that we made the right decision in choosing it.”

Find out more about FLSmidth on Twitter, Facebook, and LinkedIn.