Cloud Workloads: Can the Cloud Support EDA?

By classifying workloads based on their resource requirements and usage patterns, you can decide whether they suit the cloud.

Resource Requirements

Cloud workloads fit into categories based on resource requirements:

• A general computing workload runs on the cloud's default configuration without requiring specific computation. Web apps, servers, and data storage containers all fall into this category.
• A CPU-intensive workload contains many concurrent users and requires a high level of computing power. Examples of these cloud workloads include multiplayer online games and deep learning applications that perform processor-intensive operations, such as video encoding, big data analytics, and 3D modeling.
• A memory-intensive workload requires a large amount of processing power and memory. This category includes distributed databases, caching, and real-time streaming data.
• GPU-accelerated workloads, such as speech recognition, self-driving cars, navigation systems, and computational fluid dynamics, have exceptionally high processing requirements.
• Storage-optimized workloads include in-memory databases, highly scalable NoSQL databases, and data warehouses.

Usage Patterns

We can also classify cloud workloads by usage patterns:

• A static workload has fairly known resource requirements, demand, and uptime. Static workloads can include core enterprise services like customer relationship management, enterprise resource planning, or email.
• For periodic workloads, traffic spikes at certain times of the day, week, month, or year. A bill payment tool or a tax and accounting software are both periodic workloads. 
• With unpredictable workloads, traffic grows exponentially with little predictability. Cloud auto-scaling can handle such spikes by dynamically adding instances when needed.

Due to the CPU-, memory-, and storage-intensive nature and unpredictability of EDA workloads, chip companies should seriously consider using the public cloud. This model offers increased flexibility and scalability. 

It is essential to make the transition to the cloud thoughtfully. The key to unlocking the value of the cloud is to take a strategic approach. You should review each current deployment to fully understand compute and memory footprints and cost basis and dependencies. You also need to identify the workloads that will benefit from the public cloud. 

You should not rely on standard cloud instances for EDA workloads simply because they are easily accessible. To achieve optimal performance, the workload must match the appropriate cloud instance. One of the key metrics to use when choosing the instance is the memory per core ratio needed for your workload. Typically, front-end simulation or library characterization tools require less memory to core while backend applications, such as static timing analysis or physical verification, require larger memory to core instances. 

In addition, the cost of infrastructure and EDA licenses to run the workloads needs to be considered.