Google FeedBurner is phasing out its RSS-to-email subscription service. While we are currently working on the implementation of a new system, you may experience an interruption in your email subscription service.
Please stay tuned for further communications.
Get email delivery of the Cadence blog featured here
At the recent CadenceLIVE Americas, most of the sessions were pre-recorded with a live Q&A at the end. One exception to this was Big 3D FEM Simulations: Cloud or On-Prem? FEM stands for "finite element method" and is part of the technology that underlies tools such as the Clarity and Celsius solvers.
This was a discussion between:
Note that remarks below [in brackets] are my comments or clarification, not what someone on the panel said. Also, although I've written the post as a conversation, this is not a verbatim transcript of the panel session, that would be two or three times as long. It is edited down to the points the panelists were making. Hopefully, I haven't misrepresented anything.
Max opened the proceedings with a softball across home plate. "Is the cloud for everyone? Is it related to company size? Can small companies use it?"
David (AWS): it is hard to find an enterprise of any size not using the cloud today in some way. Anyone doing rapid scaling of simulations should be considering the cloud. I’m old enough to remember when there were special-purpose workstations for word processing and different ones for engineering. So I think it is for everyone.
Max: Having an engineering workstation on your desk was cool, although it had less power than your wristwatch today. But it was cool at the time.
Stephen (Cisco): Cloud is an economical solution for many people. If you are a massive company, it is easy to have access to lots of software packages and servers, but if you are smaller it is hard to justify having one license of everything. In some parts of the world, having access to high-end machines is maybe hard. I’m happy I have a really big LSF farm with high-end machines. But I can leverage into the cloud. If I can leverage the cloud to share licenses, that is nice. From a security perspective, small companies don’t have the same expertise as people like Cisco or AWS.
Vivek (Molex): We're lucky we have our own on-prem farms, for but smaller companies cloud offers a lot more. And within our own usage, a hybrid model is a great way to proceed in the future.
Max: I didn't have a cloud option in my day, but by the time you got to a resource someone was already using it. Sometimes you need a fail-safe.
Frank (Cadence) The smaller and mid-sized companies see a lot of value, especially for things like 3D FEM that are very compute-intensive. Also these are often only needed at certain points in the design cycle. So there is the option of going full cloud for smaller companies, and hybrid for the larger ones. But it is still evolving as to how to best utilize it. Having all the complexity removed from use for compatibility, etc is great, from networking, to hardware limitations, to OS compatibility. Cloud lets the designer focus on what he or she wants to do and not have to become an IT whizz.
Sherry: I remember at Intel using Daisy workstations, then it was huge Unix workstations, then PCs, laptops, and now the cloud. Which raises the question of security. So here's the question: how do we keep our data safe when using the hybrid approach?
David: There is a broad recognition not just in security-critical industries, not just electronics but financial services, pharmaceuticals, and more. Cloud has made it easier to create a high-security environment even compared to on-prem with the biggest enterprises. The evolution of computing platforms has been fun to watch. I started on PDP11, and now the thin client has come back again. The thin client world has come back again. You don’t need to have local data if you don’t want it. In terms of hybrid environments, there are a lot of solutions by storage vendors, compute vendors, cloud vendors like AWS. Every customer is different: legacy, multi-site…there are solutions out there and we can bring in specialists, of course.
Max: When people first started talking about cloud my reaction was that doesn’t seem very secure. But now I feel that things are more secure than on-prem, especially with small and medium and even large companies, it is turning out their security is full of holes.
Frank: The cloud solutions we are working on with AWS has security built-in. Encryption, dedicated chambers locked out from outside, chambers for the applications that are separate.
Stephen: Would you mind describing some of the things that you do for upload and download. How do you keep “my special design” safe?
Frank: We use TLS [transpot layer security] encryption.
Sherry: Question from Q&A. EDA availability in the cloud seemed to consist mostly of batch products. What about the interactive approach?
Stephen: We don’t use a lot of EDA in the cloud, we can quickly have nice containerized solutions. A small company can spin up several EDA tools.
David: Remote desktop is quite prevalent in many organizations. The "engineering workstations" has already migrated to the cloud. Highly interactive tools work really well. We have teams at Annapurna Labs [an Amazon subsidiary that designs chips in Israel] that do all their work this way. You need to choose an AWS region that is nearby, but things like massive multiplayer games are more graphically intensive than EDA tools. It does work well.
Frank: One possibility we are exploring to have tool on your local machine, and the execution engines are in the cloud. Open the tool and do your setup, and once you press "run" it goes to the cloud. That is one approach that is attractive with latency and graphic requirements. Alternatively, you can have everything on the cloud, the front-end and the engine.
Max: I always think about things as an engineer. From engineering side, there are a bunch of engineering challenges like engineer vs IT, engineers vs other engineers, engineers vs management. What should I be saying to my management? How often will we be using it? If you are talking to management how do you address these questions?
Vivek: Short answer “it depends”. There is no answer that fits all scenarios. Key as a user, I look at three things that we use to make the decision. Ease of use, security, and cost. It all translates to ROI. Then it is easy for management to decide what model is the sweet spot for that particular company
Frank: One thing that is unique about our solver is that it does take advantage of distributing the design across a lot of resources, so in the cloud, you can optimize. If you want to add more resources then you can scale that up. You can try it for a week and see if it is viable and then proceed. There is no huge investment up front, there is lots of flexibility to tweak the requirements and see how things work out, and then you can take that to management.
David. Any business is trying to optimize against constraints such as licenses, IT resources, time pressure, and so on. Getting flexibility by cloud is important for business optimization. You have constraints on licenses, time pressures, engineering resources, and so on. Cloud allows you to optimize these other aspects, especially trying to reduce risk, and especially when you have multiple projects to co-optimize.
Stephen: If you don’t have a plan, you might be worried about the large bill. Like for cellphones, if your plan is not unlimited and you have teenagers. So how do I get the flexibility up to some point, but also avoid the need to explain a big overrun to my VP?
Frank: I get that this is more of a sales and marketing decision, the solution is very scalable both for time and number of cores. If you just need 32 cores for a week, use EDAcard for that, and if you need a month of thousands of cores we can address that, too.
Sherry: Clarity 3D Cloud we announced last week. We are being very flexible, and Clarity Cloud is “for free” and you get the added boost of the cloud when you need it. We are trying to be very flexible, but Clarity Cloud you get for free, just click the button. The follow-on question to that one is that as we get more and more things going to the cloud does that make it more or less expensive for us? David, this one is crying out for you.
David: Obviously scale brings advantages of cost. We do pass on those cost reductions. Performance cost keeps going down. But as cloud has more and more scale, there are more and more choices in the cloud. The scale of cloud means you can have what might previously have been exotic hardware (GPU, FPGA) available, or even just huge massive servers. There are no EDA-specific services in cloud, beyond high-performance CPUs and storage.
Sherry: Vivek, I see you on both sides of the coin here, you do analysis, but then you have customers who might want to analyze a connector on a board.
Vivek: Our customers want to simulate their own PCB with our connectors, and that needs a lot of memory and a lot of computational power. We are set up for that on a regular basis inside Molex, but a customer might only need it once. So it is definitely an advantage to see cloud usage. In terms of cost, when does it make sense for a customer to use the cloud? It does depend on size of models.
Frank: If you have design cases that are very large but not something you solve on a day-to-day basis, then you need a lot of resources just a few times per year. So doesn’t make sense to scale your on-prem data center to that limit. If you are using a field solver for hours and days every day, and the problems are not too large, then on-prem might be a good solution, especially if you already have a data center. One thing unique about having EDA company do the cloud is that the field solver is constantly being enhanced, so as we adjust the algorithm we know the hardware tradeoffs (more memory? more CPU? more GPU? And so on). Clarity is highly parallelized but not so memory-hungry. But it benefits from a lot of CPUs, there are things we can tweak internally that would be hard for a customer on their own.
David: Diversity of choice and optimizing everything together is really hard to do on-prem. On cloud, you can continually optimize between thousands of different choices, pricing models, hardware architectures, x86 vs Arm, FPGA, GP, and so on. On-prem you have to design for a sort of worst case and everything else has to share that setup.
Frank: At Cadence, we spend months and months optimizing these configurations.
Stephen: Is it hard? Do i need to pick the right region? The right CPU for the meshing that you need to solve? Does that need to be tweaked for my particular case? Maybe I don’t have a lot of licenses. How does that work?
Frank: This is software-specific so I can just talk about Clarity. Some portions are not parallelized, and portions that are massively parallelized. You need to dedicate a large server to handle those serial parts. For the parallel part, we can use a lot of nodes and scale it up from 32 cores to 64 cores to 128 cores and on. But just what are the optimal settings is something that we spend a lot of time investigating.
David: Software users have the capability to do similar optimization. There is a choice between 4GHz x86 or maybe Arm to get a much lower cost of compute but perhaps at a lower throughput. How you trade that off is complex trading off turnaround time, cost, throughput.
Max: Thinking as an engineer arguing against an IT department who never seem to help much. Although they are very knowledgeable they don’t know the optimal configuration for any particular hardware. How would you argue against IT guys who want to keep everything in-house?
David: My experience is that IT is a partner. Many organizations have dedicated IT orgs for EDA that are more knowledgeable than their generic IT teams. Cloud is good for the careers of the IT teams, too.
Max: You used to have to be in the office to do things, and now I can imaging going to the beach with a laptop, launching the simulation in the cloud, and then going off with the family.
Frank: Not only is it easy to run a simulation in the cloud, it doesn't require multiple steps to get there. But you can close your laptop, disconnect, come back, reconnect. We don’t need anyone to do anything that causes them to do something they don’t want to like change security protocols, or do something against company policy.
Max: We're out of time. Thanks everyone.
[I don't seem to be able to provide a link that goes straight to the video of the session. If you are registered for CadenceLIVE Americas, then you can go here and scroll down a couple of screenfuls to find the description of the session and a link for replay. If you are not registered, you can still register and watch this on replay.]
Clarity 3D Solver Cloud product page.
Clarity Cloud video with Ben Gu (3 minutes):
Sign up for Sunday Brunch, the weekly Breakfast Bytes email