One of the new things at DAC this year is the Design Infrastructure Alley. The Alley is an initiative from the ESD Alliance and the Association for High-Performance Computing Professionals to highlight the information technology infrastructure, and how it is changing. To find out more, I had a call with Derek Magill, the executive director of the HPC Pros, along with the ESD Alliance's Bob Smith and Paul Cohen. I pointed out that the recent ESD Alliance CEO panel turned out to be mostly about cloud. You can read my posts about that evening at The ESD Alliance CEO Panel: Forecast Very Cloudy and CEO Panel: Cloudy with No Chance of Meatballs . “How did infrastructure alley get started?” I asked Derek was the initial driver. He has been involved in supporting engineering at TI, Qualcomm, and other places for over two decades, and is involved with the license management user-group. He thought it would be good to talk about more than licenses. Everyone’s EDA infrastructure is based on 80s and 90s technology, but the world has moved on a lot. Derek, and the ESD Alliance, took the idea to the DAC organizing board, and Infrastructure Alley was born. At DAC, and in the EDA industry in general, we talk a lot about chip designers and tools, but never about what goes on behind the scenes to keep everything reliable. Cloud appeared in EDA in the early 2000s (although we didn’t use the “cloud” word, or “SaaS”) and then largely disappeared. It took a long time for hosted design services to develop, and it doesn’t scale very well. If a company wants a server and some tools for a year, that is easy to set up. If they want 1000 servers for a month, that’s simply not viable financially with the current business model. From an IT perspective, Derek said, it is 100% true that the tools available in, for example, AWS are superior to what companies have internally. They don’t just supply scalable servers, but patches, physical security (guards), upgrade paths. Over the last year or so, it seems to me, there has been a change of opinion in the design world since when AWS or Microsoft say their security is better than on-premise server farms, everyone believes it. A couple of years ago, the standard opinion was that no semiconductor company would let their crown jewels out of their own buildings. Then I think everyone noticed that all their HR and finance systems had left the building already. Another big deal is 3 rd party IP. You can’t just take your own design, and some EDA tools, and run in the cloud. You need PDKs from the foundries, and almost always IP from several 3 rd parties. But this requires coordination. A missing piece, like a DRC rule deck, is a fatal flaw. A more subtle issue Derek pointed out is that EDA customers really don’t want the EDA suppliers like Cadence to know precisely how many licenses are used and how much over time. It potentially is a negotiating weakness. But realistically, you can’t get scalability without giving up some secrecy about this. Scalability is the big challenge. Today we are still rooted in the earlier generation of technology with MFS v3, flex-based licensing, ClearCase, LSF job management. This is inferior to what the cloud offers. Derek said: Today, if my LSF server goes down, I have large numbers of people sitting around, and kit sitting idle. But cloud is built on the assumption that everything will fail sometimes, and an individual node is not significant. The infrastructure semiconductor engineering currently runs on is very brittle and not resilient. It is also much better at keeping track of what is going on. On the old Posix running on NFS, if someone is readying and writing files there is no log. Whereas ObjectStore in AWS has everything built in: traceability, security, scalability. The 5 big topics in infrastructure are: Grid computing Storage management Data security License management Cloud The Alley is not only for IT administrators, but people who do flow development, glue-ware, and develop the EDA tools. The industry needs all the legs of the stool to move from an environment with 30 years of technical debt, and start to pay it down. It is a transition that will take time, since many companies have millions of dollars invested in hardware infrastructure, but eventually that will need to be refreshed and the economics will change and “just” adding a couple more servers to an existing on-premise datacenter will not look so attractive. Of course, for a startup, creating an on-premise datacenter is already unattractive compared to renting infrastructure-as-a-service. Find out More Infrastructure Alley is on the 1 st floor of the exhibits. There is also a “Design on Cloud” pavilion with speakers and panelists from 10.30 to 5.30 Monday to Wednesday. They will be talking about all these issues. On Monday, there is a networking event for anyone interested from 7-10pm at the Thirsty Bear brewpub (which also, slightly incongruously, serves tapas). More beer and wine on Tuesday with the reception from 5-6pm by the Design on Cloud pavilion in the exhibit hall. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧