Dynamic resource allocation and accounting in a digital marketplace.
Automating resource allocation and multi-metric accounting in a federated digital marketplace.
Kings Distributed Systems is a Canadian company deploying the Distributed Compute Protocol (DCP), a web platform that aggregates excess computing power from underutilized devices and digital infrastructure and makes it available to researchers and innovators. Their Compute API allows users to trivially express parallel workloads, e.g. Advanced Research Computing, AI/ML, blockchain, mathematical finance. The Protocol automatically distributes those workloads for computation.
A researcher’s disease modeling job that would have taken 34 days on a single computer only took 3.9 hrs using DCP on EOSC and CENGN infrastructure. Furthermore, the researcher only needed 5 lines of code to go from running their job on one computer to deploying it across 1000. The simplicity of the tool is a major advantage for researchers. We were able to collect performance metrics from the Workers and from the DCP Scheduler while under load. These metrics helped identify bottlenecks in the system design. Those bottlenecks have now been addressed and the DCP Scheduler can now handle 150,000 connected nodes, and these new Schedulers can be scaled horizontally to accommodate millions of connected Workers. We are hoping to benchmark the new Worker and Scheduler redesign on EOSC infrastructure to see how it compares to previous performance metrics.
How they used EOSC-HUB SERVICES
We provided a DCP Worker Debian package that was deployed at two locations: 8 CPU cores at Catalonia, Italy, and 8 CPU cores in Strasburg, France. EOSC staff deployed the workers, configured the “compute credit” payment accounts, and monitored credit balances and resource utilization. Feedback on performance and ease-of-use was provided throughout. DCP looks forward to testing the next iteration of the platform with EOSC staff and infrastructure.
THE VALUE PROPOSAL
The Distributed Compute Protocol streamlines massive and cheap computing. DCP eliminates the complexity of configuring environments, version control, dependency management, resource allocation, and accounting. Clients can leverage DCP to lower their cloud computing spend and to accelerate prototyping and discovery.
How EOSC-HUB helped
EOSC-HUB helped in both direct and indirect ways. Directly, EOSC-HUB provided a testbed and IT support to deploy and benchmark our solution. Performance metrics and user feedback from these tests played a central role in helping us redesign certain components of our software solution, leading to 100x improvement in performance. Indirectly, our work with EOSC-HUB provided a positive credibility signal, allowing us to attract paid pilots and investment. We are very grateful to EOSC-HUB, and certainly hope to keep working with them