Real-time Systems
The Real-time Systems group addresses three properties concerning embedded systems related to timeliness in Real-time Systems, Energy Efficiency in embedded and microgrid systems and security of IoT systems with a Blockchain and Secure Elements approach.
Real-Time Scheduling
A real-time system is a reactive, deterministic system whose proper functioning depends not only on the logical accuracy of the calculated data, but also on the respect of certain temporal constraints. In scheduling theory, a real-time system is usually formalized by a set of tasks to be executed on a platform. A scheduling algorithm must determine the order of tasks that optimizes the use of the platform while respecting the timing constraints.
We have studied the performance of fixed priority scheduling algorithms at task level and job level as well as the cost of scheduling for sporadic task templates and acyclic task graphs.
We were also interested in real-time scheduling for energy harvesting systems (Energy Harvesting System). We have proposed a task model for this problem and a scheduling algorithm for which we proved optimality in the class of fixed priority algorithms in the case and where the system is composed only of energy consuming tasks and proposed sufficient scheduling tests in a more general case. We have also studied probabilistic approaches for the modeling and analysis of real-time scheduling problems, in the case of mixed criticality systems and for time sensitivity analysis of real-time systems.
We have also studied different interconnection systems and networks such as Network on Chip (NoC) and real-time Ethernet Time Sensitive Network (TSN) in order to characterize the worst end-to-end response time. A first contribution is an adaptation of the trajectory approach traditionally used in avionics networks (AFDX) to calculate the worst case response time of TSN (AVB) networks.
Finally, we have considered the problem of certified UAVs (drones) systems for surveillance applications. We propose scheduling algorithms to schedule different Virtual Machines (VMs) having their own scheduler by a certified hypervisor (pikeOS from Sysgo). This two-level scheduling requires defining the scheduling at level 1 (hypervisor level) and at level 2 in each VMs. At level 2, we establish the schedulability conditions for sporadic tasks scheduled with a fixed priority scheduling at VM side for an arbitrary VM scheduling pattern at level 1. Then we propose new scheduling solutions at level 1 based on combination of strict periodic scheduling and Proportional Fainess (Pfair) scheduling. These algorithms have been proposed in the context of CEOS project (https://www.ceos-systems.com/en).
Energy Efficiency
We address the problem of energy supply and demand side management in industrial microgrid context. Due to increased energy costs and environmental concerns such as elevated carbon footprints, centralized power generation systems are restructuring themselves to reap benefits of distributed generation in order to meet the ever-growing energy demands. Microgrids are considered as a possible solution to deploy distributed generation which includes Distributed Energy Resources (DERs) (e.g., solar, wind, battery, etc).
We are interested in addressing energy management challenges in an industrial microgrid where energy loads consist of industrial processes. Our plan of attack is to divide the microgrid energy management into supply and demand sides.
In supply side, the challenges include modeling of power generations and smoothing out fluctuations of the DERs. To model power generations, we propose a model based on service curve concepts of Network Calculus (NC). Using this mathematical tool, we determine a minimum amount of power the DERs can generate and aggregating them will give us total power production in the microgrid. After that, if there is an imbalance between energy supply and demand, we put forward different strategies to minimize energy procurement costs. Based on real power consumption data of an industrial site located in France, significant cost savings can be achieved by adopting the strategies. We also study how to mitigate the effects of power fluctuations of DERs in conjunction with Energy Storage Systems (ESSs). For this purpose, we propose a Gaussian-based smoothing algorithm and compare it with state-of-the-art smoothing algorithms. We found out that the proposed algorithm uses less battery size for smoothing purposes when compared to other algorithms. To this end, we are also interested in investigating effects of allowable range of fluctuations on battery sizes.
In demand side, the aim is to reduce energy costs through Demand Side Management (DSM) approaches such as Demand Response (DR) and Energy Efficiency (EE). As industrial processes are power-hungry consumers, a small power consumption reduction using the DSM approaches could translate into crucial savings. We focus on DR approach that can leverage time varying electricity prices to move energy demands from peak to off-peak hours. To attain this goal, we rely on a queuing theory-based model to characterize temporal behaviors (arrival and departure of jobs) of a manufacturing system. After defining job arrival and departure processes, an effective utilization function viii Abstract is used to predict workstation’s (or machine’s) behavior in temporal domain that can show its status (working or idle) at any time. Taking the status of every machine in a production line as an input, we also propose a DR scheduling algorithm that adapts power consumption of a production line to available power and production rate constraints. The algorithm is coded using Deterministic Finite State Machine (DFSM) in which state transitions happen by inserting a job (or not inserting) at conveyor input. We provide conditions for existence of feasible schedules and conditions to accept DR requests positively.
To verify analytical computations on the queuing part, we have enhanced Objective Modular Network Testbed in C++ (OMNET++) discrete event simulator for fitting it to our needs. We modified various libraries in OMNET++ to add machine and conveyor modules. We also setup a testbed to experiment with a smart DR protocol called Open Automated Demand Response (OpenADR) that enables energy providers (e.g., utility grid) to ask consumers to reduce their power consumption for a given time. The objective is to explore how to implement our DR scheduling algorithm on top of OpenADR
Blockchain for IoT Systems
In the context of modern connected world, the concept of atomic data transfer/transaction has been completely redefined. Traditional distributed databases solve the issue of data safety through classical Atomicity, Consistency, Isolation and Durability (ACID) properties. However, the complex issue of transaction security remains difficult to address. Introducing blockchain (BC) as distributed database partially solves the problem but another issue arises i.e., BC’s modeling and evaluation. For e.g., what parameter values are ideal, is the selected blockchain framework compatible.
We address the problem of dimensioning of BC using graph theory. With binomial distribution and preferential attachment models, we model the underlying BC P2P network to reduce topology control overhead while ensuring high flexibility, fast reconfigurability, connectivity, small diameter and clustering. Next, to reduce the no. of connections per peer, we establish ideal bounds on outbound and inbound connections that still guarantee P2P network feasibility and connectivity using r-out digraphs. For an already developed BC framework, we evaluate its applicability through topology mapping. We demonstrate the efficiency of our approach using our BTCmap framework applied to Bitcoin and present its real captured snapshot.
BC alone cannot holistically secure transaction as it only guarantees data immutability whereas in most scenarios, the data has also to be secured at point of generation and usage. Further, BC has high overhead and cannot penetrate to lower levels in a system. To mitigate this, we propose the use of Secure Element (SE) to establish “root of trust”, following the “secure by design” paradigm.
Using these two technologies as the base of our proposed decentralized system, we apply it to three disparate fields. In Smart Grids, we address the problem of designing of distributed marketplace using the concepts of blockchain, SE, applied smart contracts, escrow accounts. We also address the issue of large data storage on blockchain for DR and centralization in DR allotment by designing a decentralized autonomous bidding system. We also propose a fair and efficient DR allotment algorithm whose execution time is less than 1 minute for more than 20k participants.
In IoT, we apply our SE BC Stratagem (SEBS) to solve the pressing issue of holistic data security in resource constrained devices i.e., securing data at all 3 points viz., generation, storage, and usage while maintaining very low overhead and improving performance. We also address the niche issue of verification of blockchain data where a remote device which receives data from blockchain through an intermediary, does not have resources and online connectivity to verify it. By proposing SEOVA's double signature algorithm using the technology of SE, we successfully solve this issue without compromising on security and privacy.