Sunday, September 25, 2011

EDA clouds?

Resource management has been around since clusters started in production, and even earlier in main-frames. There is something that it should happen in a cloud, invisible for the user, but in resource intensive applications this a different discipline and know-how. Grid computing, the predecessor cloud computing, could not exist without a resource manager.   Runtime Design Automation, (RTDA), a 16 year old privately held company from Santa Clara California, explains:
Our products are designed to help you get the most of your resources, whether they be computing hardware or expensive software licenses. Our goal is to maximize your efficiency while reducing waste.

RTDA has  a workload scheduler,  NetworkComputer™ and  three more complementary  products.
  1. FlowTracer automated the design flow. In real life, all resource managers handle "jobs" a single piece of work which is part of a design flow process normally done with scripts or GNU Make. FlowTracer can speed up ten times the design cycles
  2. WorkloadAnalyzer™ : Workload Automation Simulator.
    Taking a job scheduler's workload (NetworkComputer™, LSF™) along with the configuration of the compute farm (hardware & software resources) as input, a workload simulation can uncover advanced analytics of a particular compute farm configuration.
  3. LicenseMonitor™ : Monitor Software License Utilization
The last product is perhaps the main reason why resource managers still exists. Prices in hardware and networking dropped drastically, but the design automation, like EDA (Electronic Design Automation) have licenses costing in six digits per year per user and seven digits  per year for entire design teams

The entire EDA industry was not ready for applying the concept of Cloud Computing in resource managers, because of the silent opposition from large EDA ISVs whose business model was to make money from few expensive licenses and not many more lower cost licenses priced as utility..

The company does not do marketing on ongoing basis.  The word of mouth generated a worldwide customer base. The company has offices in Europe, Israel, Japan. Resource management is not a growing market, largely because the high cost of licensing prevents the widening of the market.

The need of a cloud in EDA is an ongoing effort and Synopsis published a white paper on this, very convincing that this is the next step to go. You can download it from here (requires registration). Synopsis even implemented a verification AWS cloud solution.

This online service is not for everyone. As Synopsys stated in their White Paper:
Rapidly increasing gate size and complexity in the ASIC/SoC industry has fueled an exponential growth in verification demands. Unfortunately, server technology has failed to deliver the performance increases necessary to keep up. A few years ago, for example, a directed test simulation at the register transfer level (RTL) on a 10-million ASIC gate design with a 10Hz clock targeted to operate at 1GHz required 100,000,000 seconds, or 1,157 days (3.1 years) to simulate one second’s worth of real-time operation. Things really slowed down when the rest of the system and software were added. Gate-level simulation on the same design would be an order of magnitude slower.
Moreover "Even if computers today were doubling in performance every year, they still could not keep up with this exponential verification gap"
  • Even companies that can afford large server farms and the best verification technology face a major challenge, namely the variability of verification workload. Typically, the earliest stages of verification focus on individual blocks and subsystems, which are relatively small compared to the entire project. The server farm works at optimal capacity and even test queues are manageable. But as noted above, performance of individual tests drops precipitously as multiple subsystems are integrated and the project transitions to full ASIC and eventually system-level simulation. Individual tests are then lengthy, and queues grow large. Only through careful project planning and an advanced verification methodology can the project complete on time with a reasonable level of confidence that the design will work.  
Post a Comment

Blog Archive

About Me

My photo

AI and ML for Conversational Economy