Skip Navigation

DRUID-NET aims at the following goal and broad key objective:

Following the NFV/SDN paradigm, DRUID-NET separates the flow of information into control and data plane. At the lowest layer, the IoT-enabled application are deployed, and the generated workload (data flow) can be offloaded for further processing at the above EC layer, which provides essential virtualized services. The DRUID-NET framework collects information (control flow) about the status of the computing and network infrastructure at the EC level in order to create workload-resource profiles, update the performance model for every application, and realize the feedback control mechanism for the resource allocation and simultaneously implements a resource-aware control strategy for the CPS to be controlled (control flow). This holistic approach allows the application’s dynamical modelling taking into account various contextual information. Furthermore, the controller co-design treats the resource allocation algorithms as application components in the virtualized services. With this capacity, the DRUID-NET framework focuses on the following objectives

  1. Workload profile estimation: Going a step beyond from the pertinent literature, which only considers average and general traffic characteristics of the IoT, DRUID-NET aims to differentiate and categorize the requirements of different IoT applications using appropriate data analytic and mathematical models. In particular, we classify and categorize the IoT applications by leveraging the transmission patterns, the spatial and temporal correlation of the traffic, as well as other traffic related characteristics such as the frame size distribution, and the burstiness of the traffic of the IoT applications. The novelty of this approach is that we create prediction mechanisms to treat the dynamics and uncertainty in the corresponding traffic profiles, where each predictive mechanism will target specific categories of IoT applications with similar requirements and characteristics to define what type, the size, as well as the time and the location of the requested resources.
  2. Performance Modeling: Contrary to current approaches that provide empirical static models, we aim to develop formal, realistic and dynamic traffic and resource models applicable to emulate the generated traffic from various IoT applications. To this purpose, the DRUID-NET project will adopt hybrid dynamical models that have the capacity to include several performance metrics (i.e. state variables) and resources as control parameters (input variables). This type of modelling takes into account in a single formulation the various contributions of the diverse objectives and constraints to the performance/cost. This framework will moreover allow to discover the tradeoffs between accuracy, complexity of representation and real-time feasibility of the resource allocation strategy. Furthermore, the chosen framework will be capable of capturing structural changes interpreted as discrete jumps in the dynamics, e.g., user mobility, change in wireless protocols and topology, addition/removal of edge servers.
  3. Resource allocation: DRUID-NET resource allocation mechanism aims to develop a joint communication, computing and storing virtualization paradigm, considering the problems of simultaneously (i) allocating storage, computing and communication resources, (ii) modifying network topology/ protocol and (iii) structuring the EC data centres. Following two distinct approaches, the first one is oriented towards solving multi-objective optimization problems fast, that will in turn provide an optimal operating point for the communication network, and the computing and storage allocation in the edge/cloud servers. Then, the second approach gives emphasis on the dynamic behaviour of the resource allocation. Utilizing the dynamic hybrid model obtained, we develop system-theoretic analysis methods and stabilizing controllers, ranging from Lyapunov-based optimization to reachability analysis and model predictive control. These algorithms will be designed be practicable, and guarantee feasibility and performance specifications, such as robustness to rapid changes in the workload, resource availability, and unwanted network phenomena.
  4. Co-design of controllers and resource allocation algorithms: The DRUID-NET framework allows to encapsulate, compare, and subsequently alter the impact of the several non-idealities and it is expected to have a large impact on future control applications, where resources must be used parsimoniously, in balance with the constraints and the overall objective. Our research addresses the so far untouched challenge, of designing controllers that address a mixture of unwanted phenomena by changing the provisioning of the resources to the control algorithm, if this is deemed necessary. The new generation of controllers will be made possible by the merging of two sets of hybrid models, namely a) the performance model having as internal variables performance metrics of the infrastructure and as inputs the resource distribution and utilization, and b) the process model having, for example, variables related to position, orientation, and velocity of mobile agents, lighting conditions, room temperature, mode of operation of sensors etc.