Calibre Siemens EDA and AWS working together to accelerate chip design in the Cloud

Q&A  
 
Note: 
1. Thank you all for the questions during the webinar! Below is the transcript of the questions and answers.
2. If more support is needed for any of the issues you are facing, please open a support case at  https://support.sw.siemens.com/en-US/support-case/open

 

Q: Can I use LSF in the cloud instead of SLURM?

Yes - AWS supports several commercial schedulers including IBM LSF, Univa UGE, and Altair, as well as open-source schedulers including SLURM, SGE, and OpenPBS.

 

Q: How did you come up with the right virtual machine types and storage for the Flight Plans?

A: We are constantly running benchmarks of virtual machines in the cloud.  Some of these benchmarks are internal, but many (like the ones you have seen here) we do in collaboration with mutual customers on their design data.  We take the lessons learned from these exercises (not just on virtual machine types but also networking, storage, and more) and this is what we encapsulate as the flight plans that we share with our cloud customers. 

 

Q: How does AWS ensure data and IP security? On the cloud as well as in transit to/from the cloud?

A: Security and Compliance is a shared responsibility between AWS and the customer. AWS is responsible for protecting the infrastructure that runs all of the services offered in the AWS Cloud. This infrastructure is composed of the hardware, software, networking, and facilities that run AWS Cloud services. Customer responsibility will be determined by the AWS Cloud services that a customer selects. This determines the amount of configuration work the customer must perform as part of their security responsibilities.

This shared model can help relieve the customer’s operational burden as AWS operates, manages, and controls the components from the host operating system and virtualization layer down to the physical security of the facilities in which the service operates. The customer assumes responsibility and management of the guest operating system (including updates and security patches), other associated application software as well as the configuration of the AWS provided security group firewall. Customers should carefully consider the services they choose as their responsibilities vary depending on the services used, the integration of those services into their IT environment, and applicable laws and regulations.

From an implementation standpoint, encryption is the primary mechanism to ensure data and IP security on AWS, both in transit as well as at rest. Customers can choose to bring their own encryption keys and manage them on AWS using our KMS service or use AWS managed keys. We also provide identity and access management service to implement fine grained access control to data I.e. who can access and do what with the data on AWS. There are several other networking based security control mechanisms starting from the transport layer all the way to the application and presentation layers.

 

Q: On-prem grids are managed by our CAD infrastructure team, whereas the cloud is managed by our IT organization. What would the operating model look like as EDA workloads start migrating to the cloud?

A: It varies from customer to customer, but in many cases, we're seeing the CAD teams getting very hands-on in the cloud and working with us to optimize their tool flows.  The important thing is that collaboration between the CAD and IT teams is key to successful migration into the cloud. 

 

Q: Do you have a way to enable spot instances for the flows that you described?

A: This is flow-dependent, but for example, in Calibre we have a way to manage the use of spot instances in a way that is fault-tolerant, so the flow is not disrupted if spot instances disappear. 

 

Q: How much lag is there for GUI-driven applications?

In our experiments, we used AWS NICE DCV for GUI-driven applications and found it to function well and snappy in terms of latency, even when the application is hosted in a geographically distant AWS region. GUI performance can be dependent on the application, the amount of data transferred, and the remote settings.  Work with your Siemens EDA team if you encounter any issues.

 

Q: Are there any kernel-level settings that I need to run efficiently in the cloud?

A: Recommended kernel-level settings differ by product, and your Siemens EDA support team or field applications team can help you find the right ones for your flow and environment.  For some flows, there are kernel settings that we recommend in addition to the ones that we identify for on-premises environments. 

 

Q: How can I tell what are the best virtual machines to use?

A: "Best" will depend on which specific tool and flow you are using.  Please refer to the Cloud Flight Plans for guidance related to your specific workload. 

 

Q: How does licensing work in the cloud?

A: The required licensing setup differs depending on which tool you are using.  We recommend that you work with your account team to determine the best licensing scheme for your specific situation.            

 

Q: Who should I contact for more information?

A: You can always request more information through our Siemens EDA Support Center. Alternatively, reach out to your account team or field applications engineer who will know the right people to put you in touch with.

 

Q: How can I get started with Questa in the cloud?

A: Please work with your account team to get the best possible ramp-up for any of our Siemens EDA products into the cloud.  We will be releasing Questa workflow-specific Flight Plans later this year to ensure efficient cloud scaling.

 

Q: What is the typical AWS spin up time when I click the Run button in Calibre?

A: Typically, a few minutes, but it depends on how well you've primed the instances with the right AMI and the right settings to launch new instances.

 

Q: Is there any other reason behind FSx for Lustre over EFS other than IOPS?

A: It's primarily about IOPS.  FSx and EFS are POSIX-compliant.  AWS also provides integration between FSx and Amazon S3 (AWS object store) which is used by customers to transfer data from the on-premises environment to AWS.  Different tasks and workflows may benefit from different combinations of shared and local storage with container support.  Flight plans will provide guidance based on our testing.

 

Q: Are you still recommending ?6*.* instances now that ?7*.* instances are available, i.e. m7a.32xlarge seems like a better choice.

A: Our Calibre Manufacturing team has done some benchmarks with r/m/c7a instances which indicate that they are an excellent choice.  The instances currently offered have less memory per physical core than the r/m/c6i instances, and this may become a limiting factor for Calibre nmDRC/nmLVS/PERC runs on big designs that need more memory per core. 

 

Q: What’s the recommended data bridge between AWS and our data center?

A: It depends on the volume and the frequency of the data transfers that you plan to do. If you are 100% in the cloud, then of course the the bandwidth requirement is low. But if you plan to do a lot of data transfer, e.g. at a petabyte scale, then we would recommend considering what we call direct connect. This is a fiber link between your data center and the AWS backbone that AWS can work with your telco provider to set up and they can scale that pipe to as high as 100 GB. But you can also set up site to site VPN if the bandwidth requirements are low.

 

Q: Are Siemens cloud licensing prices equivalent to our on-prem prices?

A: Please talk to your Siemens EDA account team for questions on licensing. 

 

Q: Do we need to rearchitect our design verification flow to work optimally in AWS?

A: It's our goal to make your workflows in the cloud resemble as closely as possible your workflows in your on-premises environment, and that is part of the reason Siemens EDA and AWS are working closely together on these Cloud Flight Plans. 

 

Q: Are AWS instances connected effectively to storage hosts to match on-prem performance? 

A: AWS has several options for storage.  Elastic Block Store (EBS) is what we recommend as the attached storage to a particular instance.  AWS has various configurations of EBS that you can choose from to meet your performance requirements. AWS also has instances that come with NVME storage. So if you go into the AWS EC2 page you will see memory-optimized instances that comes with large amounts of NVME storage. So AWS can configure both the specific virtual machine instance as well as the choice of storage to meet or exceed what you have in your on-premises environment.

 

Q: How do we tell the EDA tool that we want to simulate using ParallelCluster (slurm)?

A: Most of our Siemens EDA tools have built-in settings that enable job submission to SLURM parallel clusters (as well as support for other queue managers like LSF). 

 

Q: Given that the cost of an AWS instance is linear, do you have any graphs showing cost savings based on using larger instances?

A: Generally, for any given class of virtual machine an instance with twice the number of cores will cost you twice as much. So, if you need that many cores, then we recommend that you go with a small number of big machines rather than a big number of small machines because (a) you don't have to worry about availability and (b) it's going to cost you the same.