Does IC design have a future "in the clouds?" Yes, according
to panelists at last week's Design Automation Conference - but selectively,
over a period of time. As attractive as cloud computing is,
there are still technology challenges and tradeoffs, and the EDA licensing
model for cloud computing has yet to be resolved.
The panel followed
a keynote address in which Bernie Meyerson, vice president of innovation at
IBM, predicted that cloud computing will become a "dominant" technology within
the next 5 to 10 years. He showed how IBM is using a private cloud for its chip
design work. There was enthusiasm in the panel, too, but panelists also gave
full consideration to cloud computing challenges and limitations. The result
was a very balanced, informative and thought-provoking two-hour panel
The panel was organized by Andreas Kuehlmann, director of
Cadence Research Labs, and moderated by Raul Camposano, consultant and the
recent CEO of Xoomsys, a SPICE simulation startup that used cloud computing. Panelists
included Rean Griffith, U.C. Berkeley; Deepak Singh, manager of business
development for Amazon Web Services; Paul Leventis, director of design
automation at Altera; James Colgan, CEO of Xuropa; Samuel George, services
group director at Cadence; and John Chilton, director of marketing at Synopsys.
Rean Griffith is a co-author of a seminal paper on cloud
the Clouds, a Berkeley View of Cloud Computing." It's thus fitting
that he started the discussion by offering some definitions. In brief:
The cloud is the sum of datacenter hardware
and software. There are public clouds, private clouds, and hybrid clouds that
combine both public and private. Key advantages of the cloud are 1) the
appearance of infinite computing resources, 2) elimination of up-front
commitments, 3) ability to pay for resources on a short-term basis, 4)
economies of scale, 5) higher utilization of resources, and 6) simplified
Cloud computing is the combination of Software as a Service (SaaS)
with utility computing. Within the cloud are multiple layers, including
Infrastructure as a Service (IaaS) and Platform as a Service (PaaS). Top two benefits
are 1) elasticity of resources and 2) risk transfer (the cloud provider is
responsible for capacity planning, not you).
What's Good About EDA in the Cloud
As you can imagine,
Amazon's Deepak Singh had lots to say about the benefits of cloud computing.
"It is very empowering to sit at a command line and be able to operate an
entire infrastructure with hundreds of servers," he said. With message-based
computing, he noted, the cloud will dynamically scale to meet the user's
resource requirements with little or no human intervention.
Singh said Amazon Web Services - a large public cloud --
is built on top of core principles including availability, durability,
scalability, elasticity, security, and effectiveness. Concerned about downtime?
Singh said Amazon S3 data storage is designed to provide "11 9s" (99.999999999%)
durability of objects over a given year, because objects are stored in multiple
data centers. If you have 10,000 objects, you'll lose one every 10 million
James Colgan, CEO of cloud computing provider Xuropa, is also a strong advocate. He noted
that Cadence has successfully been using Xuropa for about a year for pre-sales
evaluation and demos of verification IP. Prospective users don't have to
procure an evaluation license, line up IT resources, and download and install
the IP and associated simulation tools before trying it out - and people at
Cadence don't have to get on a plane.
Cloud computing, Colgan said, "is a great opportunity to
lower overall costs for the EDA industry and to deliver great value to design
engineers. As we continue our way along the technology adoption curve, we'll
learn and adapt the business models and the licensing, and we'll solve the
technical problems. Our job at Xuropa is to make it easy for engineers and EDA
vendors to take advantage of what is really a huge, massive paradigm shift."
Samuel George of
Cadence cited the "promise" of cloud computing, including infinite capacity,
dynamic access to compute resources, no up-front commitments, and a "pay as you
go" pricing model that matches demand. He noted that achieving this promise
will require an economy of scale, as well as standard (but configurable) use
Today, George said, some
companies employ both a central WAN and a cloud farm. Interactive work is done
on the central WAN, while batch work is done on the cloud farm, which is remote
from the user. "We take advantage of these two things internally and offer them
to our customers," he said, noting the Cadence Hosted Design
Solutions offering. What typically goes onto the cloud farm? Applications
like physical verification, DFM checks, circuit simulation, and logic
John Chilton noted
that Synopsys, Cadence and other EDA vendors have offered SaaS applications for
some years, but "today's cloud is different." He said that "EDA always utilizes
the highest performance available compute platform. It now looks like computing
is moving to the cloud. We in EDA had better figure out how to move our
The Dark Lining Behind the Silver Cloud
So what are the
challenges? There are obvious concerns about data security whenever data goes
across a firewall, but cloud providers such as Amazon provide extensive
assurances about security. Answers do not come as easily when it comes to
software licensing. Altera's Paul Leventis is skeptical about cloud computing,
and that's one of the reasons. "Even if I suddenly had access to a million
computers, cost elasticity is not supported by [EDA] licensing models today,"
he said. "EDA doesn't lend itself well to a SaaS model given the way most EDA
tools work today."
Leventis said that
EDA software is much more expensive than the hardware it runs on, and predicted
that even if EDA moves to the cloud, he won't save money because EDA vendors
don't want to reduce their revenues. Another drawback: EDA flows are heavily
customized, and how do you move a customized flow into the cloud? But Leventis
said he would like to use cloud computing to run massive regression tests in
parallel, for infrequent access to "big iron" computing, and for infrequent
access to "boutique tools" that it doesn't make sense to buy.
some "tradeoffs" for cloud computing. These include a reliance on remote
systems, the fact that data is off-site, reduced flexibility in design flows
(because of the need to fit into a more standardized model), and pricing that
varies based on the service level. Speaking of services, he noted that the current
SaaS model "needs to be very specifically configured for your needs. Right now
this does not work out of the box."
Chilton had a list
of questions. How should I think about security? How do I handle multi-vendor
flows? How do I get on and off the cloud? Which applications are a good fit?
What's the business model? As to the latter question, he remarked that "customers
have to believe it's fair, and we have to believe we have a business model that
won't put us out of business."
This panel provided
a lot of food for thought. I came away thinking that cloud computing is a
logical and inevitable step for many EDA and IP applications, but we need to go
into it with an awareness of the questions and challenges it poses. The
licensing model question is difficult, but I'm sure there are creative solutions.
The DAC panel was an excellent start to a dialog that needs to continue and
needs to involve the entire IC design ecosystem.