In 2002, Sun Sparc stations running Solaris was the standard workstation for engineers and researchers in the enterprise sector. These machines, while expensive, provided a solid Unix environment for hardware design, e.g., ASIC layout or DSP algorithm simulation.
Filip Körling was at the time supporting the engineers and researches for a DSP simulation department for an enterprise customer. The task was to provide a well-working and efficient simulation environment.
One of the first improvements was to connect the existing Sun Sparc stations into a compute cluster, running the LSF Platform cluster software. This removed bottlenecks and gave a much improved simulation performance as jobs were distributed to all the workstations in the department.
At this stage in the play, a new actor appears on the stage.
Enter the Mighty Penguin, also known as Linux. By now, the major EDA vendors had Linux on the radar and were starting to provide Linux support for their tool chains.
A prototype project was created where one ugly pizza box-size PC, equipped with a dual Pentium IIIs was acquired. Filip, together with Anders Westrup, managed to get the DSP simulation software to run on the machine. Regression tests were then run of characteristic simulations to verify correctness and measure performance.
While the hardware was less than attractive, the results had a strong appeal. The Linux machine provided roughly 10x the simulation performance at 1/10 of the cost compared to the Sun Sparc stations. And, it gave correct results.
A most solid business case, if you ever saw one.
Next, the department approved a purchase of 20 more Linux machines, and together with Platform LSF cluster software, a new compute cluster with 10x performance of the old was born.
As the news spread in the company, other departments also wanted better performance and started acquiring their own Linux clusters. However, these clusters were isolated, thus creating a new bottleneck; if even if one department had free capacity, it was not available to other departments.
Filip, together with Fredrick Rybarczyk, wrote a proposal to the Head of R&D management where the main point was to pool all machines together in a common cluster. The proposal was accepted and company wide compute cluster was created.
The rest, as they say, is history.
Today, Linux is taken for granted and most EDA companies today have hundreds, if not thousands, of Linux machines, connected as distributed systems and providing companies’ main compute power.
Read more blog posts from Stickybit