Do you expect widespread EDA usage in public clouds in 3 years, 8 years, or never? That's the question that was put before attendees at a Design Automation Conference (DAC) panel June 7. But first came a spirited discussion that revealed much about the motivations, advantages, and challenges of electronic design in cloud computing environments.
The panel was titled "Could Computing and EDA Forecast: Sunny Skies or Storm Clouds Ahead?" It was moderated by EDA veteran Raul Camposano, CEO of Nimbic, an EDA startup that's implementing signal-integrity analysis through cloud computing. Panelists were as follows:
The session started with a half-hour tutorial given by Singh on Amazon Web Services. He noted a set of "core principles," including on-demand access, pay-as-you-go, security, elasticity, and programmable infrastructure. He spoke of the "shared responsibility" between customer and Amazon for data security (for more on this topic, see http://aws.amazon.com/security/). Finally, he noted a key advantage for EDA, access to scalable computing resources.
Private Cloud Brings Efficiency
IBM's Anderson spoke about the success that his company is experiencing with a private cloud for IC design. "Any IBM designer can work on any project anywhere in the world," he noted. IBM is now supporting twice the number of designers on the same amount of money it spent on IT seven years ago, Anderson said, and that's not the only advantage - "what surprised us is how much efficiency improved."
"I don't think cloud computing is at all new to EDA," Bruggeman said. "Cadence has been providing cloud-based solutions for a decade." These solutions include EDA tools on the Internet, available in the late 1990's; secure VCAD chambers for customer design data, initiated in 2002; and finally Hosted Design Solutions, a more recent Software as a Service (SaaS) offering.
What's been learned from this experience? Bruggeman cited three points. First, the technology is here today to overcome challenges like security, scalability, and performance. Secondly, cloud computing is not a way to get cheap EDA tools, although it can reduce the cost of ownership over time. Third, cloud computing in the future will call for collaboration by multiple partners in the ecosystem.
Chilton remarked that "cloud computing is an inevitable move, so there's no sense in sticking our heads in the sand." But what problems should it solve? The toughest one right now is verification, he said, and the opportunity is peak demand, such as needing 1,000 CPUs on a weekend to make a Tuesday deadline. Chilton noted that Synopsys now provides on-demand, metered access to its VCS simulator through the Amazon cloud.
Ten Years or Three?
After discussing the differences between internal clouds, private clouds, and public clouds, Chian noted that cloud computing is following a rather typical technology adoption curve that runs in 10-15 year cycles. "We believe the commercial reality of EDA in public clouds is 8 to 10 years away," he said. "On the other hand, the private cloud is emerging. It is in place."
EDA will come to public clouds, Chain said, but a few "considerations" need to be dealt with first. These include issues surrounding risk and compliance (such as data security and retrieval), flexibility (customization and operational control), and timing (different services will move to the cloud at different rates). In the meantime, he said, a "hybrid" model will combine attributes of public and private clouds.
Madrona's Gottesman, who introduced himself as an early stage investor who is "wrong more than right," had a very different view. "8 to 10 years is ridiculous," he said. "I think it will move much more quickly." He then presented six reasons why IC design will move to the cloud within 3 years, as follows:
Camposano offered another reason why public clouds will take over most computing. It's a lot more cost-efficient for one provider to buy a million servers than for 1,000 people to each buy 1,000 servers, he said.
It's a Cultural Thing
When it comes to EDA in the clouds, the first topic that normally comes up is security. But the real issue is culture, Bruggeman said. "In the EDA space, we design the way we design because that's the way we design, and we've done it for years," he said. "We've learned how to solve the security challenges but we haven't learned to overcome cultural challenges. That puts the 3 year adoption at risk and makes 10 years more likely."
This made sense to me. At the end of the panel, when Camposano asked the audience whether they expected EDA public cloud adoption in 3 years or 8 years, I joined about two-thirds of the audience in voting for 8 years. Significantly, however, I didn't notice anyone voting for "never."