So, CAE engineer friend, having doubts about this cloud thing when it comes to professional data? Me too. Well not really, I must admit things are starting to clear up a bit. Let me tell you why.
When you come to think of it, I even wonder why I am writing this blog. We have all become so used to sharing our latest pics and videos in the cloud, that extending that behavior to professional data should be a non-question. But, it isn’t for us of the CAE community: we have – or in my case, had – doubts about the whole thing…
This blog article will address the case of the puzzling slow adoption of cloud-based solutions by the CAE community, the reasons thereof and will expose some innovative solutions recently made available.
According to a recent survey by RightScale, a leading US-based cloud computing consultancy (source: http://assets.rightscale.com/uploads/pdfs/RightScale-2015-State-of-the-Cloud-Report.pdf): “88 percent of organizations are using public cloud while 63 percent are using private cloud,” “55 percent of enterprises report that a significant portion of their existing application portfolio is not in cloud, but is built with cloud-friendly architectures.” But, they also said “68 percent of enterprises run less than a fifth of their application portfolio in the cloud.”
Hum, hum. A majority of organizations are using the cloud, even if applications are on the slow side when it comes to taking full advantage of the technology. Putting a magnifying glass over our small CAE world, it is worth noticing that on-premises HPC clusters have been for many years a viable option. Given that the distance between on-premises HPC and CAE-in-the-cloud seems small, one would think that adoption rate of cloud for CAE would be high.
And yet, it is a strange world. Based on informal research amongst current Ceetron customers and analyses of available tool packages in the market, the answer is no. I confirm what my marketing colleagues found: the answer is DOUBLE no, based on my personal 20-year experience as an FEA/CFD software professional.
So, CAE engineer friend, our gut feeling reflects the true world: CAE community is far behind on cloud usage compared to other fields. The natural question that pops up is of course: why? What are the dark forces (only a few days to wait!) that prevent such a highly technical community from taking advantage of the highly technical cloud?
In the world of CAE, cloud computing is to a large degree understood to be simulation on a remote hardware infrastructure combined with data storage on a central file system. Data modelling, preprocessing, post processing and visualization are mostly done on desktop computers. These work processes are characterized by a high degree of interactive activities, something that the cloud infrastructure is not well suited for.
Traditionally, CAE-in-the-cloud is therefore based on one of the following two approaches:
- The interactive processes (modelling, preprocessing, result interpretation, visualization) are being performed locally on the desktop. The simulation is run on a cloud resource. Input data and output data are transferred back and forth between the desktop and the cloud resource.
- The complete workflow from geometry modeling to post processing and visualization is executed on the cloud and all services are accessed through a remote desktop front-end or one or more dedicated web browser interfaces.
In approach (1), we could dub “isolated interaction”, a lot of data shoveling is taking place in the background. In a typical FE simulation, the amount of data involved in the preprocessing is fairly limited; say tens, or at the most, a few hundreds of megabytes. This data is comfortably conveyed on the network if necessary. But, after the FEA or CFD simulation is executed, the dataset has put on some weight… In fact, much more than some: tens of megabytes may have become tens of gigabytes, which is less easily moved around. Hence, this approach strikes its limit when simulation datasets grow in size. Of course, hardware improves as well, and a complex and advanced backbone can help mask the basic issue. But “isolated interaction” workflows and FEA/CFD simulations will not live happily ever after: models sizes are growing (see earlier articles https://blog.ceetron.com/2015/05/28/technology-prediction-for-big-3d-typical-fea-and-cfd-model-sizes-to-be-used-in-the-cae-community-in-2020-and-2025/ and https://blog.ceetron.com/2015/07/13/future-model-sizes-in-the-fea-and-cfd-community-predictions-from-interviews-with-industry-insiders/)
In approach (2), a.k.a. “remote interaction”, all work processes – including interactive ones -are accessed AND executed on the cloud computer. The cloud computer requires both GPUs and CPUs, to handle both the graphics involved in the interactive processes and the equation solving involved the simulations. At Ceetron, we know how much user experiences are highly dependent on graphical performance, and our blunt statement is that the comfort of “remote interaction” suffers from variable network bandwidth.
In a nutshell, the limitations of remote interaction are at least one reason for the slow pick-up of cloud technologies in the CAE communities. The data transfer issues remain, as the only alternative is isolated interaction. Think about all those minutes wasted watching progress bars fill… annoying isn’t it, my CAE engineer friend?
Stuck? Luckily no longer. In comes WebGL, a sort of OpenGL for web browsers – the stuff you need to render 3d graphics. It has been around since 2011, but only very recently developed into a de facto graphics standard on all client devices ranging from desktop computers to handheld devices. Indeed, it may reasonably be predicted that all major CAE vendors over the next few years will rewrite their interactive software applications to utilize WebGL.
What’s in it for me? Because the interaction is handled locally, WebGL-based solutions provide the user comfort remote interaction cannot, and at the same time limit data transfer to the strict minimum during interaction, i.e. only what is required and when it is required.
And this where my colleagues and I at Ceetron come into play in the cloud frenzy. Ceetron has worked closely with world-class research organizations and industry partners for a period of almost ten years to create web- and cloud-based technologies for visualization of and collaboration on CAE models. Some recent research projects include: Direct visualization in the cloud (2010-2014), Mobile Forging Workbench (2012-2014), and 3D visualization of engineering simulation data on the web (2015-2017).
At the time we started Ceetron Cloud, the choice of WebGL technology was a bit of a gamble. Side-note: I do have a few good poker players among my colleagues, and this hand is proving to be a good one. Just some days after its first release, more than 300 models have been uploaded. The success of the system relies also on the fact that all Ceetron products act as enablers and that their unique send-to-cloud feature can be bolted on to any solver or simulator offering.
Ceetron Cloud has indeed resolved some of the issues implied by either “isolated interaction” or “local interaction” by creating a WebGL-based client-server visualization engine that combines compute resources on the cloud HW and utilization of the local GPU of the desktop computer (or any other device for that matter). What differentiates Ceetron Cloud from previous architectures based on the same idea is a server that generates highly optimized, compacted 3D objects – the minimum amount of data required for interaction – that are streamed to the client for local rendering. The local renderer works independently from the server, and hence, provides the user with an interactive experience as close as possible to that of a native desktop application. In effect, the network bandwidth does not influence on the interactivity of the client process.
In sum, by combining the computational power of the cloud with the WebGL graphics rendering capabilities of the local GPU, we have taken a partial, but crucial step towards a successful cloudification of CAE applications.
If not convinced, join the movement at https://cloud.ceetron.com.
Preliminary benchmark results are also highly compelling, though we need more time to conduct an engineering-wise methodologically-sound performance comparison.
In a follow-up blog post, my colleague Fredrik Viken, CTO in Ceetron, will provide a technically-oriented blog post about the technical implementation of Ceetron Cloud. We will also in separate blog post revert to the issues of security, both for private and for public clouds, and formal benchmarking. Finally, NAFEMS will host a conference on HPC adoption for SMEs with CAE needs in Manchester in February 2016 and I think that we will be there.