We often focus on the largest high performance computing deployments but a bulk of real enterprise HPC workloads happen in far less sophisticated environments, albeit with applications that still the need scalability, performance, and efficiency the big centers can provide.
Here’s a sense of how deep the divide is between the big centers doing large-scale modeling and simulation for engineering applications and those with big HPC needs but less advanced resources to support such goals. Almost one-fifth of Ansys’s survey base say that their most important, frequent simulations have to run overnight, taking more than 9 hours to complete.
The reasons for this can’t possibly access to compute resources as might have been in the pre-cloud days. That means the problem with HPC access is much more local. According Ansys director of strategic partnerships, Wim Slagter, “people still struggle with HPC; they’re hesitant to scale up and migrate some of their simulation workloads from workstations, for example, to HPC clusters. They are reluctant because of the IT complexity and many lack the HPC support and appropriate HPC hardware.” He adds that scaling and configuring an HPC cluster for engineering simulations is “not so easy and these users are compute-bound, so to speak.”
Based on the in-depth survey, Ansys found that cloud is becoming an increasingly important deployment model for speeding simulations. As many regular readers here know, cloud is an obvious choice for plenty of enterprise workloads but the path to cloud in HPC has been very slow due to a range of factors, ranging from interconnect and network performance to moving large data volumes to one of the trickiest barriers, a historical lack of porous software licenses (although that is quickly becoming a problem of the past). What all this means is that when cloud is used for HPC applications, the tooling is not as developed on the part of ISVs or even the cloud providers for the unique needs of scalable HPC codes on remote hardware.
From the surveyed base, “more than a quarter indicated that public cloud or an ISV-managed solution could reduce turnaround time limitations on simulations,” which might not sound like much given the rest of the wider world. Even five years ago, seeing a quarter of HPC workloads operating in an ISV-run or public cloud would have been surprising. But what drove some of that momentum? Covid. With remote work becoming the norm and lack of organizational workstations or small clusters came a new comfort with the cloud—or if not comfort, a forced alliance.
“Many engineers are basically reluctant to expand the use of simulation and HPC. From our last survey (but also previous ones we conducted), it became clear that the lack of IT hardware and support is the biggest barrier to expanded use of simulation,” Slagter says. “And the need for ROI, business value and technical proof are perceived as barriers to expanding or adopting HPC. At Ansys, our strategy for HPC aims to remove these barriers. We are basically addressing the top factors that must exist before customers invest in new HPC resources.”
Some highlights from the Ansys report include the following:
- Half of small to mid-sized enterprises are using workstations exclusively versus slightly over one third for large companies.
- In 2020, 31% of respondents were using a combination of mobile/desktop and a departmental server/cluster. That figure was 23% in 2014.
- According to the current survey, 13% of respondents are running mostly on a company’s HPC datacenter but do the pre- and post-processing on their local workstation.
- Only 3% of respondents conduct pre-processing, solve and post-processing on their company’s HPC datacenter.
- In addition, they are increasingly running their simulations on higher numbers of compute cores and relying on both on-premises and cloud HPC resources to address their simulation challenges.
- End users are relying on larger numbers of CPU cores and parallel processing capabilities. According to the survey, 18% are using more than 36 cores and 9% use more than 132 cores. In 2014, just 10% were using more than 32 cores, and only 3% of respondents were using more than 128 cores.
“In our first cloud survey, improving business agility was ranked near the top of business and IT benefits. So the ability to scale up and down as needed, was the top justification of cloud computing costs for companies of all sizes,” Slagter says.
“Through our most recent survey, the adoption of remote collaboration and data management tools is a top-of-mind technology priority. This is related to the pandemic and the widespread practice of social distancing. Remote working became the new reality of many businesses! Needless mentioning, the role of IT is increasingly getting pushed by engineering to deliver more and better IT systems for simulation, and cloud is clearly an ideal platform to make this happen.”
“Customers want to see proof or evidence of HPC benefits for their own simulation models,” Slagter says.”
To this end Ansys is letting users play with some numbers to evaluate different deployment options. This can certainly be useful to those with HPC workloads that don’t match the Ansys workload set around modeling and simulation for engineering and design. Users can upload their Ansys model so that we can run their model on a small HPC cluster from HPE. They will then receive a time comparison versus the configuration of their own desktop computer. In this way, they will know precise time savings possible for their own model.