Device-Edge-Cloud Intelligent Collaboration framEwork (DECICE)

The DECICE project was created in response to an EU Horizon research and innovation programme calling for the development of a cognitive cloud system. Such a system should be able to integrate cloud, HPC and edge resources and optimize their usage. Furthermore, machine learning based schedulers will automatically optimize the placement of workloads across the heterogeneous hardware landscape managed by it. As part of this project, challenges must be solved regarding the integration of distinct hardware architectures into a unified platform as well as building a process for gathering metrics and finding near optimal scheduling decisions in a reasonable amount of time while also scaling to thousands of nodes across multiple data centers and countries.

• Dr. Julian Kunkel
• Jonathan Decker
• Stefanie Mühlhausen
• Felix Stein
• Michael Bidollahkhani
• Aasish Kumar Sharma

• University Göttingen (Coordinator)
• GWDG
• E4
• KTH
• HLRS, University of Stuttgart
• HUAWEI Technologies Düsseldorf
• Forschung Burgendland
• SYNYO
• Marmara University
• TOP-IX
• University Bologna
• The Numerical Algorithms Group
• BIG TRI

The DECICE project aims to develop a unified compute continuum for HPC, Cloud and Edge resources. This system should be able to administer such a heterogeneous hardware landscape while providing tools for administrators and users to utilize the resources. Furthermore, the system should integrate an AI scheduling system to optimize the placement of workloads for performance and energy consumption. This AI scheduler is to be trained on monitoring data including CPU, memory, and network usage, as well as, application performance. As a benchmark for the capabilities of the AI scheduler serves a heuristic scheduler that uses filters and prioritization rules to determine workload placements.

While the scheduling framework will be portable, a prototype for the system will be built on top of Kubernetes. Kubernetes itself is only designed to handle cloud resources, making it necessary to utilize additional projects to integrate Edge and HPC resources. The integration of Edge resources will be handled by KubeEdge and the integration of HPC resources will be handled by Volcano. Patches and new feature implementations that are developed in response to technical shortcomings of these software projects will be sent back upstream to the open source communities.

The code produced for the DECICE project will also be released as open source and any research papers will be published as open access.

The architecture of the DECICE project can be seen in the following graphic. The DECICE Model and DECICE Manager include components that will be developed while the underlying compute plane shows what software will be used for the prototype.

The DECICE project received funding from the European Union's Horizon 2022 research and innovation programme under grant agreement No 101092582.