UK cloud management company YellowDog builds multi-regional supercomputer in the cloud in 33 minutes


LONDON, November 23, 2021 – Yellow dog, a UK cloud workload management company, built a cloud supercomputer in just 33 minutes. The announcement marks a decisive step forward as, for the first time, thousands of servers across Europe and North America were able to work together seamlessly on a single issue. This frees large IT tasks from the constraints of a particular system or data center, dramatically reducing the cost of high performance computing and accelerating critical innovation, from months to hours.

This breakthrough means the cloud is now even more accessible and scalable for key research areas, such as drug discovery and artificial intelligence. While high performance computing has always played a pivotal role in advancing important areas of exploration, funding supercomputers can take years, with significant investments required both to build and maintain the infrastructure. This prevents many organizations from realizing its benefits.

YellowDog has introduced a new reality, where growing computing needs can be met instantly in the cloud. In sixty-five minutes, he ran a popular drug discovery application on 3.2 million virtual processor cores, as a single cluster. Provisioning a million preemptible instances took just seven minutes.

YellowDog also offers organizations the choice of accessing supercomputer performance with minimal environmental impact. With significant sustainability benefits, the technology not only provides the exact resources needed, when they are needed, but it also enables the selection of data centers based on renewable energy production. This means high performance computing becomes much more efficient and IT tasks can be placed where clean energy is most available. This ability to move where the sun is brightest, or the wind strongest, has never been achieved before.

“This number of simultaneous cores, stretching across the Atlantic, is a huge step forward. This reduces the time, cost and, most importantly, the carbon impact of researching some of the most important societal issues, ”said Simon Ponsford, CEO of YellowDog. “Our platform, thanks to its ease of use, allows organizations to focus on the problem itself, not being an expert in harnessing the potential of the cloud. The domains of the possible are no longer linked by the coordination of computing resources, but by the vision of the human mind.

“In the past year alone, we’ve seen not only a continuation of the global pandemic, but also significant displacement caused by extreme weather events,” Ponsford continued. “The ability to solve these complex and multifaceted problems relies on the collection, processing and generation of information from large data sets. This demonstrates the growing importance of high performance computing in efforts to advance the fields of science, engineering, and technology. “

About YellowDog

YellowDog, founded in 2015 in Bristol, UK, helps businesses scale in the cloud like no other. The company’s cloud-based workload management platform is used around the world to accelerate compute-intensive applications.

The YellowDog platform removes prohibitive barriers to entry, so organizations of all sizes can address some of the most immediate threats society faces, including the global pandemic, multiple and severe weather events, and research into ‘renewable energy. YellowDog works with cloud partners including Amazon Web Services, Microsoft Azure, Google Cloud Platform, Oracle Cloud Infrastructure and Alibaba Cloud.

About the race

To achieve this execution, YellowDog launched 46,733 preemptible instances, using 24 AWS EC2 fleets and 8 instance types, provisioning a total of 3.2 million concurrent vCPU cores to run a drug discovery application. The job consisted of over 200,000 separate tasks, spread across Europe and North America.

Overall, the race was completed in 65 minutes. It took only 7 minutes to provision 1,000,000 cores. Not only was it fast, but actual instance usage followed provisioning. This meant that the instances were prepared according to the ones being used, thus maximizing the efficiency of use.

Source: Yellow Dog


Comments are closed.