FogStore – A Data Management Platform for Geo-Distributed Fog Environments

• Applicant: Professor Dr.-Ing. David Bermbach
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: from 2018 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 415899119

Today’s applications are typically cloud-based. However, the emerging Fog Computing paradigm, i.e., using cloud and edge resources but also small- to medium-sized data centers in the network between edge and cloud at the same time, promises additional benefits in terms of quality of service (QoS) for some applications. Especially, emerging application domains such as the Internet of Things (IoT), autonomous and interconnected driving, future mobile networks (5G), or eHealth can benefit from or even heavily depend on these QoS improvements. The increasing degree of geo-distribution, however, requires rethinking system and application architectures. Data management systems in particular – both of the relational and NoSQL kind – are ill prepared for this degree of geo-distribution and other characteristics of fog environments. In essence, this means that fog developers currently have to implement data management tasks at application level.With FogStore, we aim to close that gap. Namely, the core results of the project will include:(i) A data management system that supports arbitrary, dynamically changeable data distribution policies as defined by applications while dealing with aspects such as QoS constraints or failures.(ii) An application-side middleware component that uses lightweight monitoring for replica selection in case of moving application clients.(iii) A manager for predictive replica placement that actively moves data between edge locations, anticipating future application requests. For this, the middleware will analyze past application access patterns and application-provided hints through time series analysis methods as well as machine learning to predict future application access positions. Thus, a replica can be provisioned close to that position before the data is accessed.(iv) A novel benchmarking framework designed for geo-distributed and coordinated execution of several parallel workloads. Such a framework can – and will in the proposed project – be used to evaluate performance and other QoS properties of fog data management systems but also more generally of data management systems for geo-distributed environments, e.g., as in cloud federation.

• DFG Programme: Research Grants