SaaS – which has its origins in the 1960s – can provide a range of functionality, including outsourcing IT and software. And it’s big business; some estimates say SaaS is now a $20billion industry.
While much of the action in SaaS has been in the commercial arena, the approach is finding wider application in engineering –and the EDA sector is seeing renewed interest.
Cadence has been offering SaaS since 2008, when it launched Hosted Design Solutions. Now IBM has now joined the party, making three of its tools available via its Cloud service.
Larry Drenan, services group director, said: “Cadence has offered SaaS since 2008, with standard Cadence software run in a managed environment and accessed remotely. As new software is introduced, it can be included in this environment.”
IBM’s Jeff Karmiol, senior product manager for HPC Cloud, said: “We’ve made the software available because we see an opportunity in a marketplace that hasn’t been well served. It’s a tool flow developed over the years for use in house.”
IBM has launched three EDA tools – library characterisation, logic verification and Spice simulation. Karmiol noted: “We’ve had requests to sell our tools and it seems there’s the right convergence of opportunities; we have a high performance Cloud and that’s the right place to run the tools.”
Christopher Porter, IBM’s HPC Cloud Product manager, added: “While we’re only launching three tools, we think they’re the ones which address customers’ pinch points.”
Porter noted that IBM has developed an end to end tool flow. “We continue to use those tools in house, so there will be quite a few tools available for release on the cloud in the future.”
Jacqi Tull, Mentor Graphics’ director of pricing and packaging and license compliance, said: “We have always expected that EDA companies would embrace SaaS, because it’s another way to get product to customers; opening up new markets and making new products available. Mentor is always looking at new technology and new ways to deliver products and the interesting thing with SaaS is that customers have the potential to see the whole design chain.”
Are all EDA products appropriate for SaaS? “There are a few point solutions which may be good for SaaS,” said Drenan, “if they require large numbers of servers and licenses and represent an isolated step in the design flow. Library characterisation and certain highly parallel types of mathematical simulation are examples.”
IBM’s Porter is more enthusiastic. “A full EDA design flow comprises around 30 tools, but that’s the complete flow. If you look at the number of jobs executed, those are handled by a much smaller number of tools.
“For example,” he continued, “as logic verification moves to smaller design rules – particularly 12nm and beyond – the amount of verification has been increasing in a non linear fashion and around 85% of all compute cycles are now consumed by some kind of verification tool. This is the kind of problem which causes a ‘pinch point’.” And so, Porter implies, this is an opportunity for SaaS.
Karmiol added: “The ability to pay as you go brings a new way of thinking for designers. They know what their throughput is, but they also need a plan to handle that. For example, with SaaS, they may be able to access more powerful computing facilities for a lower price because they don’t have to own the hardware or pay for software licenses.”
Mentor’s Tull believes SaaS will meet the customer’s needs at times, but wonders whether that will be more of a point solution. “SaaS may prove attractive in assessing technology.” she said. “It’s a way to try software with relatively low risk and this is one thing which the customer will look at.
“But it may also be beneficial when companies are looking to share the design effort. It could be a workgroup, but it could also be a third party you’re inviting to take part.”
Drenan pointed out that, despite the fact the Cloud offers seemingly limitless computing power, EDA software itself will limit performance. “We have recently introduced products that take greater advantage of compute power to reduce run time, but you should realise that not all EDA programs [suit this approach]. The scale possible on the Cloud is orders of magnitude larger than some algorithms will ever be able to exploit.”
IBM’s EDA services are offered on a Platform LSF cluster, which manages resources, schedules workload and monitors events. Jobs are submitted to the head host, then distributed to compute hosts. The larger the cluster, the more processing power is available and up to 128 compute hosts can be used. “The pay per use model makes things easier,” Porter asserted.
“Cadence helps customers by ramping their compute power when they need it for certain phases of the design, or to take advantage of new tools,” said Drenan. “But EDA companies do not have unlimited private computing power, so we will be able to take even more advantage of abundant processing when we can blend our software offerings and design environment management experience onto the public Clouds.”
Porter said there is no reason why a company with an EDA license can’t deploy their tools on the Cloud. “We already have an electronics client doing that because it can get access to things they don’t want to invest in.”
Two other issues loom when it comes to using the Cloud – security and communications. Karmiol said: “We make sure every bit is encrypted using AES. Once in the Cloud, the shared file system is also encrypted and can’t be accessed without keys. Once data is transferred out, we use a DoD standard disk wipe.”
Tull said SaaS security is a huge concern. “There are solutions, but as soon as someone is hacked, people do ‘doubletakes’,” she said. “For some, the security issue won’t be solved, but it’s part of the risk management process.”
Drenan noted that one of the first things Cadence does is to run a latency test to measure the quality of the connection. “An ISP connection that works for email and web browsing may not stand up to the higher performance characteristics of chip design,” he said. “Through experience and testing, we have learned how things like average latency, standard deviation, and dropped packets will affect the user experience.”
Tull sees some downside to SaaS. “It will be a challenge for customers with complex, multivendor tool flows where third party IP is used. For them, their tool chain can be their ‘secret sauce’. There will be times where SaaS meets customer needs, but I think it will be more of a ‘point tool’.”
Drenan concluded: “To offer a high quality model at a reasonable price, SaaS providers have to standardise. If this approach works, SaaS can be a good solution. But if the customer needs a custom methodology, customised set ups or equipment, this might not be a good model.”
