Section: New Results

Evaluation and optimal scaling of real time systems

Code analyses and advanced visualization of software in real-time

Participants : Pierre Caserta, Olivier Zendra.

This work has significantly progressed. A thorough state of the art has been realized and resulted in the submission of a paper to the TVCG journal.

The implementation of instrumentation, tracing and analyses (see VITRAIL software in this report) has been done, to provide us with the necessary basis for our experiments. This design work is the topic of another paper which we have begun writing.

Now, we have entered the experimentation phase that will allow us to implement, test and validate our new ideas.

Low-power memory placement

Participants : Maha Idrissi Aouad, René Schott, Olivier Zendra.

Based on last years work, a bibliographic paper about low-power memory placement strategies is about to be finished and submitted.

Since software developments were taking more time than expected, work in this domain has been refocused on more theoretical aspects. These consist in modeling energy consumption of various parts of the architecture, especially caches, and trying to find better placements strategies in memory.

Iterative multicriteria optimizations in critical, real-time systems

Participants : Jonathan Ponroy, Olivier Zendra.

Work in this domains is performed in the context of the ANR MORE (Multicriteria Optimization for Real-time Embedded systems) project, which involves three partners (LIP6 in Paris, IRIT in Toulouse, and us). This project aims at finding cooperative strategies to jointly improve several criteria (namely energy, code size and WCET computability) in real-time, critical systems, in iterative setting. We at INRIA Nancy Grand Est are responsible for the energy criterion.

As explained in the software related part of this report, after taking important delays in the software developments that should have been done for the project, we started on better bases and made significant progress in implementation. A join paper is beginning to be written by the project partners about the platform we have developed this year.

This experimental platform, finally provided us with the experimental setup we had directly needed and enabled us to progress on the scientific front. Currently, we have evaluated the impact of single criterion transformations. We are finishing the evaluation of bi-criteria interactions, which will allow us to tackle the next part of the project, which is the semi-automated exploration of multicriteria compromises.

Open Power and Energy Optimization PLatform and Estimator

Participants : Sophie Alexandre, Kévin Roussel, Olivier Zendra.

Work in this domain is performed in the context of the ANR Open-PEOPLE (Open Power and Energy Optimization PLatform and Estimator) project, financed since the end of 2008. For more details about the project, see the corresponding section under the Grants chapter of this report. INRIA Nancy Grand Est is responsible for the software part of the platform.

Work in this project has begun in April 2009 (kick-off meeting). We have so far finished setting up the very important infrastructure for the software part of the Open-PEOPLE platform. We are finishing expressing the requirements for the platform, in order to start the actual developments and the actual integration of tools provided by the different partners. The research work itself is also is about to begin. INRIA Nancy Grand Est is involved in peak power control issues, and memory management for low-power issues. Note that we have difficulties finding good candidates for the PhD we propose.

Real time deterministic multiprocessor scheduling

Participants : Liliana Cucu-Grosjean, Olivier Buffet [ INRIA Nancy-Grand Est ] .

We deal in this topic with deterministic scheduling of tasks on different processors; the schedule must be done such that the deadlines are satisfied. For the classical model of tasks, we give in [26] a solution based on a constraint satisfaction problem (CSP) that we prove equivalent to the multiprocessor problem. We propose two different CSP formulations. The first one is a basic encoding allowing to use state of the art CSP solvers. The second one is a more complex encoding designed to obtain solutions faster. With these encodings, we then study the resolution of the scheduling problem using systematic search algorithms based on backtracking.

Probabilistic scheduling of real time systems

Participants : Liliana Cucu-Grosjean, Dorin Maxim.

We deal here with probabilistic scheduling of real time systems with variable execution times. Since some parameters of a system can be unknown until the time instant when the activity is released or the environment can change forcing the application to adapt, we need to consider an approach able to address this type of scheduling and we investigate the use of probabilistic approaches to solve this problem. We are interested in two different problems:

• the study of priority assignment in the uniprocessor case [33] . More precisely we deal with fixed-priority scheduling of synchronous periodic task systems with variable execution times. The tasks have variable execution times given by independent discrete random variables (that we consider known) and we study the existence of an optimal priority assignment algorithm for such tasks. We provide a first result indicating that Rate Monotonic is not optimal and we prove that optimal priority assignment algorithms do exist. Moreover a first intuitive algorithm, that orders the tasks according to their probability of meeting the deadlines, is proved not optimal.

• the study of schedulability analysis in the multiprocessor case [48] . More precisely in the case of preemptive fixed-priority scheduling of periodic task systems with variable execution times we provide (naive, but efficient) improvements to existing uniprocessor analyses. These improvements are based on a discussion on the validation of probabilistic schedulability analyses. For the case of several identical processors, we give a new probabilistic schedulability analysis for global fixed-priority scheduling.

Sensitivity analysis for real time distributed systems

Keywords : schedulability analysis, distributed systems.

Participants : Liliana Cucu-Grosjean, Reinder Bril [ Eindhoven University of Technology ] , Joël Goossens [ Université Libre de Bruxelles ] .

Existing end-to-end response time analysis in distributed real-time systems, where the finalization of one task on a processor activates another task on another processor, is pessimistic. By "pessimistic" we mean that not all systems deemed to be unschedulable by the analysis are in fact unschedulable. This pessimism has two causes: (i) the existing analysis is based on best-case response times rather than best-case finalization times and (ii) those best-case response times are based on analysis for (worst-case) deadlines at most equal to periods minus (absolute) activation jitter. We present in [25] analytical means to determine best-case finalization times of independent real-time tasks with deadlines larger than periods minus activation jitter under uniprocessor fixed-priority preemptive scheduling (FPPS) and arbitrary phasing, allowing an improvement of the existing analysis. Moreover, we deal in [24] with exact best-case response times of periodic released, independent real-time tasks with arbitrary deadlines that are scheduled by means of fixed-priority pre-emptive scheduling (FPPS). Apart from having a value on its own whenever timing constraints include lower bounds on response times of a system to events, the novel analysis allows for an improvement of existing end-to- end response time analysis in distributed systems, i.e. where the finalization of one task on a processor activates another task on another processor.

Robustness evaluation for a critical distributed system

Participants : Adrien Guénard, Lionel Havet, Françoise Simonot-Lion.

In order to improve mobility in cities, new initiatives of Intelligent Transportation Systems are emerging based on free-access electric vehicles. Making the vehicles everywhere at anytime available requires a dispatching of these vehicles over a city, this could be accomplished at a large scale with platoons of vehicles. We developped a methodology to assess the performances of such urban platoon of vehicles using different platooning algorithms, taking into account a real operational architecture which means communication delays, task jitters, data sampling and infinite resources. As a result some timing properties requirements are provided for the choice of each vehicle operational architecture for the deployment of the platooning function. Secondly an analysis is made on platooning operational architecture in order to evaluate the robustness of the platoon of vehicle under transient faults leading to information losses for the control of the vehicles [30] .

Robust deployment of a real-time in-vehicle embedded middleware

Participants : Liliana Cucu-Grosjean, Dorin Maxim, Nicolas Navet, Françoise Simonot-Lion.

This study is part of the PREDIT-SCARLET project. This year we study how to evaluate the robustness of a solution given by each frame-packing algorithm developped during past years face to several transient faults (delayed signals, delayed frames, etc.) For this purpose, we have done sensitivity analyses based on probabilistic approaches [49] . Under non-preemptive hypothesis, we have considered the analysis of periodic messages with activation jitters that are scheduled on a Controller Area Network (CAN) bus. We propose two probabilistic analyses, each of them provides probability distributions of message response times where the activation jitter of messages is given by independent random variables. The first analysis gives the probability of having the worst-case response time and the second analysis the distribution of the average response time.

Scheduling of tasks on automotive multicore ECUs

Participants : Aurelien Monot, Nicolas Navet, Françoise Simonot-Lion.

As the demand for computing power is quickly increasing in the automotive domain, car manufacturers and tier-one suppliers are gradually introducing multicore ECUs (Electronic Control Units) in their electronic architectures. In [46] , we address the general problem of scheduling numerous elementary software components (called runnables in AUTOSAR terminology) on a limited set of identical cores. In the context of an automotive design, we assume the use of the static task partitioning scheme which provides simplicity and better predictability for the ECU designers with respect to a global scheduling approach. We show how the global scheduling problem can be addressed as two sub-problems: partitioning the set of runnables and building the schedule on each core. Then, we prove that each of the sub-problems cannot be solved optimally due to their algorithmic complexity. We then present low complexity heuristics and derive lower bounds on their efficiency (i.e., competitive ratio). Finally, we assess the performance of our approach on realistic case-studies.

Aperiodic traffic in response time analyses with adjustable safety level

Participants : Dawood Khan, Nicolas Navet, Françoise Simonot-Lion.

In distributed real-time systems it is crucial to ensure the temporal validity of the data exchanged among the nodes. Classically, the frame Worst Case Response Time (WCRT) analyses, and the software tools which implement them, do not take into account the aperiodic traffic. One of the main reasons for this is that the aperiodic traffic is generally very difficult to characterize (i.e., the arrival patterns of the aperiodic frames). The consequence is that one tends to underestimate the WCRT, which may have an impact on the overall safety of the system. In [32] , we propose a probabilistic approach to model the aperiodic traffic and integrate it into response time analysis. The approach we develop allows the system designer to choose the safety level of the analysis based on the system's dependability requirements. Compared to existing deterministic approaches the approach leads to more realistic WCRT evaluation and thus to a better dimensioning of the hardware platform.

Networked control systems: resource overload management using selective data dropouts according to (m, k) -firm model

Participants : Flavia Felicioni [ Rosario University, Argentina ] , Ning Jia, François Simonot, Françoise Simonot-Lion, Ye-Qiong Song.

The stability and performance of a networked control system are strongly influenced by the network delay and packet drops. We consider that late arrived sampling data are dropped, so that we only focus on the analysis of the impact of packet drop sequences on the control loop stability and performance. For any dropping sequence specified by (m,k)-firm model, and considering a simple mono-variable linear system with a proportional controller and zero control action in case of sampling data drop, we derived the stability conditions based on the upper bound of the plant state variance. It has been shown that the stability only depends on the values of m and k but not the pattern of the dropping sequence. In case of network overload, this gives much freedom to actively dropping some packets while still keeping the system stable. An analytic method to determine the optimal control gain for any given packet drop pattern is also derived, providing thus a guideline for optimal control and network resource scheduling co-design ([29] ).

Performance evaluation and optimization

Participants : Francis Comets [ LPMA, University Paris 7 ] , Liliana Cucu-Grosjean, François Delarue [ LPMA, University Paris 7 ] , Lhassane Idoumghar, René Schott.

We analysed the deadlock phenomena occuring in (real-time) distributed systems sharing common resources [10] . In our model transition probabilities of resource allocation and deallocation are time and space dependent. The process is driven by an ergodic Markov chain and is reflected on the boundary of the d-dimensional cube. In the large resource limit, we prove Freidlin-Wentzell estimates, we study the asymptotic of the deadlock time and we show that the quasi-poyential is a viscosity solution of a Hamilton-Jacobi equation with a Neumann boundary condition. We give a complete analysis of the colliding 2-stacks problem and show an example where the system has a stable attractor which is a cycle limit.

We have designed a novel hybrid evolutionary algorithm that combines Particle Swarm Optimization (PSO) and Simulated Annealing (SA) algorithms. When a local optimal solution is reached with PSO, all particles gather around it, and escaping from this local optima becomes difficult . To avoid premature convergence of PSO, we present a new hybrid evolutionary algorithm, called PSOSA, based on the idea that PSO ensures fast convergence, while SA brings the search out of local optima because of its strong local-search ability [31] . The proposed PSOSA algorithm is validated on ten standard benchmark functions and two engineering design problems. The numerical results show that our approach outperforms algorithms proposed recently by A. Abraham, A. Ishigame, S. Nakano, P.M. Thanjaraj and K. Yasuda.

The multiprocessor scheduling problem consists in finding a schedule for a general task graph to be executed on a multiprocessor system so that the schedule length can be minimized. The multiprocessor scheduling problem is known to be NP-hard, and to obtain optimal and suboptimal solutions, several heuristic based algorithms have been developed . We propose two original Tabu Search type algorithms for solving this problem [41] . Our heuristic algorithms are validated on 13 randomly generated instances. The numerical results show that our algorithms produce solutions closer to optimality and/or of better quality than the methods presented by Chauvière, Geniet and Schott.

The frequency assignment problem involves the assignment of discrete frequencies to the transmitters of a radio network, such as a radio broadcasting network. Frequency separation is necessary to avoid interference by other transmitters to the signal received from the wanted transmitter at the reception region. Here, it is of major importance to minimize the interference while at the same time using the spectrum efficiently. We developped two original distributed algorithms implemented on clusters of PCs used to solve the frequency assignment problem in the field of radio broadcasting. The first one is based on the island distributed implementation of our hybrid genetic algorithm. The second one uses a distributed cooperative Tabu Search. Experimental results show that our algorithms, applied to several instances given by TDF-C2R, lead to important time performance improvements [13] .