Cloud computing can generally support the types of workloads required by EDA tools, but when it comes to billion-gate semiconductor simulation, there's room for improvement. A recent collaboration between Xuropa, Intel and Cadence, presented at the user track at the June Design Automation Conference (DAC 2012), resulted in a server management software framework that can boost throughput by as much as 15% for billion-gate simulations.
The paper was titled "Billion Gate Semiconductor Design and Simulation and the Next Phase of the Cloud Computing Evolution." It was presented by James Colgan, CEO of EDA cloud provider Xuropa, and Naresh Sehgal, software architecture manager at Intel. Both are interviewed in the short video below. Cadence provided EDA tools and support for the experiment.
In brief, the server management software schedules workload execution according to the real-time state of the underlying hardware. It can dynamically scale available CPU, I/O, and memory resources for each workload. The framework determines optimal workload placement and cloud configurations to meet the demands placed on the hardware resources available. In the video, Sehgal uses an air traffic control analogy to describe how it works.
The experiment described in the paper was executed on an OpenStack based cloud. Execution times of the workloads were recorded both with and without the framework technology, and then compared. The resulting up-to-15% improvement in workload is significant for several reasons. Not only can jobs run 15% faster - you can run 15% more jobs. Both time-to-market and product quality improvements follow, along with lower design and verification costs.
What kinds of resources are required for a billion-gate simulation? How does the new software improve throughput? How is scheduling handled? What were the results? And what was the Cadence role in this collaboration? The video below has answers. Click on the icon to view or click here.
Richard Goering