Haut PDF Optimality and non-preemptive real-time scheduling revisited

Optimality and non-preemptive real-time scheduling revisited

Optimality and non-preemptive real-time scheduling revisited

We first embark on analyzing non-idling scheduling. The optimality of the non- idling non-preemptive Earliest Deadline First scheduling policy is revisited. Then, we provide feasibility conditions in the presence of aperiodic or periodic traffic. Second, we examine the concept of idling scheduling, whereby a processor can remain idle in the presence of pending tasks. The non-idling non-preemptive Earli- est Deadline First scheduling policy is not optimal since it is possible to find feasi- ble task sets for which this policy fails to produce a valid schedule. An optimal algorithm to find a valid schedule (if any) is presented and its complexity analyzed. This paper shows that preemptive and non-preemptive scheduling are closely related. However, non-preemptive scheduling leads to more complex problems when combined with idling scheduling.
En savoir plus

34 En savoir plus

Non-preemptive scheduling algorithms and schedulability conditions for real-time systems with precedence and latency constraints

Non-preemptive scheduling algorithms and schedulability conditions for real-time systems with precedence and latency constraints

Unité de recherche INRIA Rocquencourt Domaine de Voluceau - Rocquencourt - BP 105 - 78153 Le Chesnay Cedex France Unité de recherche INRIA Futurs : Parc Club Orsay Université - ZAC des V[r]

37 En savoir plus

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Non-Pre-emptive Scheduling

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Non-Pre-emptive Scheduling

[12] George, L., Rivierre, N., Spuri, M., “Preemptive and Non- Preemptive Real-Time UniProcessor Scheduling”, INRIA Research Report, No. 2966, September 1996. [13] George, L., Muhlethaler, P., Rivierre, N., “Optimality and Non-Preemptive Real-Time Scheduling Revisited,” Rapport de Recherche RR-2516, INRIA, Le Chesnay Cedex, France, 1995. [14] Howell, R.R., Venkatrao, M.K., “On non-preemptive scheduling of recurring tasks using inserted idle time”, Information and computation Journal, Vol. 117, Number 1, Feb. 15, 1995. [15] K. Jeffay, D. F. Stanat, C. U. Martel, “On Non-Preemptive Scheduling of Periodic and Sporadic Tasks”, In Proc. RTSS, pages 129-139, 1991.
En savoir plus

11 En savoir plus

Scheduling of parallel real-time DAG tasks on multiprocessor systems

Scheduling of parallel real-time DAG tasks on multiprocessor systems

We present simulation results of preemptive GDM scheduling algorithm and we compare the performance of DAG-Str algorithm and the Direct Scheduling approach at DAG-Level, by varying the n[r]

257 En savoir plus

Real-Time Fixed and Dynamic Priority Driven Scheduling Algorithms: Theory and Experience

Real-Time Fixed and Dynamic Priority Driven Scheduling Algorithms: Theory and Experience

To that end, our purpose is first to provide a general framework based, on the one hand, a representation of preemptive, real-time scheduling in an algebraic structure that enables us to evaluate the distance of the optimality of any scheduling algorithm ; and on the other hand, a consistent representation of the associated feasibility conditions that enables us to evaluate the number of basic operations. As a second step, considering several kinds of traf- fics, we initiate the comparison by a straight, but limited, application of our general frame- work. Our preliminary results will notably highlight, in the cases where deadlines are all greater than periods, that fixed priority schedulers (like deadline monotonic) behave as well as EDF while the worst-case response time analysis is less complex. The same observation is valid when the task sets are almost homogeneous but is in favor of EDF in the general case or when a simple feasibility analysis is needed.
En savoir plus

145 En savoir plus

Schedulability conditions for non-preemptive hard real-time tasks with strict period

Schedulability conditions for non-preemptive hard real-time tasks with strict period

Although, some works allow the computation of the exact cost of preemptions in the scheduling analysis [4], usually this cost is approximated as stated by Liu and Lay- land [1]. This approximation may lead to incorrect be- haviour during the real-time execution of the tasks or at least a waste of resources due to the WCET and memory margins the designer must take. In the same vein the over- head of preemptive scheduling algorithms is more difficult to characterize and predict than the one of non-preemptive scheduling algorithms. Since scheduling overhead is of- ten ignored in scheduling models, an implementation of a non-preemptive scheduler will be closer to the formal model than an implementation of a preemptive sched- uler. In this case, the cost of the scheduler itself could be taken into account in schedulability conditions. Non- preemptive scheduling on a uniprocessor naturally guar- antees exclusive access to shared resources and data, thus eliminating both the need for synchronization and its asso- ciated overhead. In control applications, the input-output delay and jitter are minimized for all tasks when using a non-preemptive scheduling discipline, since the inter- val between the start and end times is always equal to the task computation time [5]. This simplifies the tech- niques for delay compensation in the control design. In many practical real-time scheduling problems involving I/O scheduling, properties of device hardware and soft- ware either make preemption impossible of prohibitively expensive [6]. For these reasons, designers often use non- preemptive approaches even if the theoretical results do not extend easily to them [7].
En savoir plus

10 En savoir plus

Analyse of real-time systems from scheduling perspective

Analyse of real-time systems from scheduling perspective

In [ZKG + 08], the authors propose a framework for the schedulability analysis of real- time systems, where they dene a generalized model for sporadic tasks to more precisely characterizes the task arrival times. Each task is characterized by two constraints: higher instantaneous arrival rate, which bounds the maximum number of task arrivals during some small time interval; lower average arrival rate, which is used to specify the maxi- mum number of arrivals over some longer time interval. The work of [MCG13] considers systems with probabilistic execution times and probabilistic inter-arrival times. However it does not handle dynamic scheduling policies. Moreover, the method is a numerical analysis technique whose complexity is exponential in proportion to the number of sam- ples and tasks. In [TDP12], the authors propose a method to control the preemptive behavior of real-time sporadic task systems by the use of CPU frequency scaling. They introduced a new sporadic task model in which the task arrival may deviate, according to a discrete time probability distribution, from the minimum inter-arrival time. Based on the probability of arrivals, the authors propose an on-line algorithm computing CPU frequencies that guarantee non-preemptiveness of task behavior while preserving system schedulability.
En savoir plus

149 En savoir plus

Real-time scheduling of transactions in multicore systems

Real-time scheduling of transactions in multicore systems

Real-time scheduling algorithms. The classification for uniprocessor sys- tems is usually based on the priority (static or dynamic) assigned to tasks. For preemptive uniprocessor systems, Earliest Deadline First (EDF) is optimal [11]. Unfortunately, EDF is not optimal on multiprocessors either under the par- titioned or the global approaches [5], called respectively P-EDF and G-EDF. Another class of scheduling algorithms, which differs from the previous ones, is the Pfair algorithm [2]. It is based on the idea of proportionate fairness and ensures that each task is executed with uniform rate. Tasks are broken into quantum-length subtasks and time is subdivided into a sequence of subintervals of equal lengths called windows. A subtask must executes within the associ- ated window and migration is allowed for each subtask. PD 2
En savoir plus

6 En savoir plus

Algorithms for Preemptive Co-scheduling of Kernels on GPUs

Algorithms for Preemptive Co-scheduling of Kernels on GPUs

Abstract—Modern GPUs allow concurrent kernel execu- tion and preemption to improve hardware utilization and responsiveness. Currently, the decision on the simultaneous execution of kernels is performed by the hardware, which can lead to unreasonable use of resources. In this work, we tackle the problem of co-scheduling for GPUs in high competition scenarios. We propose a novel graph- based preemptive co-scheduling algorithm, with the focus on reducing the number of preemptions. We show that the optimal preemptive makespan can be computed by solving a Linear Program in polynomial time. Based on this solution we propose graph theoretical model and an algorithm to build preemptive schedules which minimizes the number of preemptions. We show, however, that finding the minimal amount of preemptions among all preemptive solutions of optimal makespan is a NP-hard problem. We performed experiments on real-world GPU applications and our approach can achieve optimal makespan by preempting 6 to 9% of the tasks.
En savoir plus

11 En savoir plus

Simulation of Efficient Real-Time Scheduling and Power Optimisation

Simulation of Efficient Real-Time Scheduling and Power Optimisation

Fig 9 Task execution with DVFS 7.5 Comparison of results We have proposed an approach which is to describe and compare the energy efficiency of various power management techniques and this combined with an EDF preemptive scheduler. Also, we could raise the energy profile of each processor in addition to the execution of the the tasks in order to give more visiblity concerning the performances of STORM to highlight the energy savings. In addition, we determined, via the simulator, the power consumed by each processor calculated for a period of 1 second. The CPU power consumption diagram of a processor shows over time its electrical power (in watts) computed according to the physical characteristics of its chip and its functioning states. Insofar, as the processors, generally, function with less possible downtime, the energy energy consumption and costs obtained for each technique of power management differ considerably. Energy saving can mostly reach 95% of the costs of the consumption compared to the use of EDF. At first glance, it should be noted that the techniques of energy management contribute contribute significantly to the reduction in power consumption, although the use of DVFS seems more efficient in terms of gain. It is about 0.05J only whereas without energy consideration it reaches 0.93J and this without negative incidence on the performance of the systems.
En savoir plus

10 En savoir plus

Modeling and scheduling embedded real-time systems using Synchronous Data Flow Graphs

Modeling and scheduling embedded real-time systems using Synchronous Data Flow Graphs

52 State of art Scheduling independent non-preemptive strictly periodic tasks A more particular scheduling problem is the one with non-preemptive strictly periodic tasks. The non-preemptive scheduling problem is known to be NP-hard computational complexity [65]. Adding the strict periodicity constraint increases the problem com- plexity. Korst et al. [80] were the first to study the problem of scheduling a set of non-preemptive strictly periodic tasks. Their work was motivated by real-time video signal processing. The authors considered this problem on a minimum number of processors [79]. They showed that the problem is NP-complete in the strong sense, even in the case of a single processor, but that it is solvable in polynomial time if the periods and execution times are divisible. Thus, they proposed an approximation algorithm based on assigning tasks to processors according to some priority rule. Besides, they proposed a necessary and sufficient condition for the schedulability of two strictly periodic tasks. Later, Kermia et al [70] proposed a sufficient schedulability condition that generalizes the previous condition for a set of tasks. They imposed that the sum of the tasks execution times is less or equal to the greatest common divisors (GCDs) of task periods. In [92], Marouf and Sorel proved that this sufficient condition is very restrictive (pessimistic). They also gave a schedulability condition for implicit-deadline strictly periodic tasks and proposed an heuristic based on this condition. In contrast, our approach is not restricted to implicit-deadline tasks.
En savoir plus

192 En savoir plus

Preemption Delay Analysis for Floating Non-Preemptive Region Scheduling

Preemption Delay Analysis for Floating Non-Preemptive Region Scheduling

† University of Rennes 1, UEB, IRISA, Rennes, France Email: {jmsm,nelis,smp}@isep.ipp.pt , Isabelle.Puaut@irisa.fr Abstract—In real-time systems, there are two distinct trends for scheduling task sets on unicore systems: non-preemptive and preemp- tive scheduling. Non-preemptive scheduling is obviously not subject to any preemption delay but its schedulability may be quite poor, whereas fully preemptive scheduling is subject to preemption delay, but benefits from a higher flexibility in the scheduling decisions. The time-delay involved by task preemptions is a major source of pessimism in the analysis of the task Worst-Case Execution Time (WCET) in real-time systems. Preemptive scheduling policies including non-preemptive regions are a hybrid solution between non-preemptive and fully preemptive scheduling paradigms, which enables to conjugate both world’s benefits. In this paper, we exploit the connection between the progression of a task in its operations, and the knowledge of the preemption delays as a function of its progression. The pessimism in the preemption delay estimation is then reduced in comparison to state of the art methods, due to the increase in information available in the analysis.
En savoir plus

9 En savoir plus

Scheduling non-preemptive hard real-time tasks with strict periods

Scheduling non-preemptive hard real-time tasks with strict periods

There are some works in the case of non-preemptive tasks with strict periods. Al Sheikh and al. study the par- tition scheduling on an IMA (Integrated Modular Avion- ics) platform where the avionic functions are strictly peri- odic. They gave an exact algorithm with excessive com- putation time, based on a linear programming formula- tion, to solve the problem. Korst and al. proved in [11] a necessary and sufficient schedulability condition for two tasks, which becomes a sufficient condition for more than two tasks as proved by Kermia in [12]. However, as men- tioned in [13], this later condition is very restrictive. In [14] Eisenbrand and al. proposed scheduling algorithms in the case of harmonic and non-harmonic tasks.
En savoir plus

9 En savoir plus

Formal Verification of Real-time Systems with Preemptive Scheduling

Formal Verification of Real-time Systems with Preemptive Scheduling

For Scheduling-TPN, we tackle the problem of the state-space explosion by a two-stage analysis. First, we pre-compute the state space of the Scheduling- TPN as a linear hybrid automaton. This first step is performed by a fast DBM- based algorithm. Although this algorithm is over-approximating, the produced linear hybrid automaton is proved to be time-bisimilar to the initial Scheduling- TPN i.e. the additional locations generated by the approximation are actually not reachable. As a consequence, the cost of the translation is fairly low. The second step consists of an exact analysis of that LHA with the HyTech model- checker. For this second step to be efficient, the number of variables (clocks) must be kept as low as possible. To this effect, the translation algorithm offers a number of reduction mechanisms and thus produces a LHA that has, in general, a fairly lower number of variables than what is required for a direct modeling as a product of linear hybrid automata.
En savoir plus

37 En savoir plus

A T-time Petri net extension for real time-task scheduling modeling

A T-time Petri net extension for real time-task scheduling modeling

2. Scheduling Extended Time Petri Nets We propose an extension for the TPN 1 that enables to take into account the way the real-time tasks of an application distributed over different processors are scheduled. At first this extension consists in adding two parameters to the TPN’s places ; we call them “p r o c e s s o r  ” and “priority”, and they correspond respectively to the allocation and the priority of the task which is associated with the place. However all places of a TPN do not require such parameters. Actually when a place does not represent a true activity for a processor (for example a register or memory state), neither a processor nor a priority have to be attached to it. In this specific case, the semantics remains unchanged with respect to a standard TPN. One can notice that it is equivalent to attach to this place a processor for its exclusive use and any priority (it does not matter in this case).
En savoir plus

15 En savoir plus

Integrated Worst-Case Response Time Evaluation of Multicore Non-Preemptive Applications

Integrated Worst-Case Response Time Evaluation of Multicore Non-Preemptive Applications

Rapport de recherche n° 8234 — February 2013 — 17 pages Résumé : Les techniques d’analyse de pire temps d’exécution (WCET) ont atteint une très bonne précision dans l’analyse de programmes séquentiels s’exécutant sur architectures monopro- cesseurs. Dans cet article, nous étendons une technique récente d’analyse WCET et l’outil associé pour permettre de calculer des estimations précises du temps de réponse (WCRT) d’applications parallèles non-préemptives s’exécutant sur des plates-formes multi-coeurs. La technique proposée est intégrée dans le sens où elle calcule en même temps les estimations WCET des fragments de code séquentiel et le WCRT global. L’utilisation de cette methode produit des estimations plus précises que les approches découplées plus classiques où le calcul du WCRT est réalisé à partir de valeurs WCET précalculées séparément pour chacun des fragments de code séquentiel. Sur 2 exemples d’applications de contrôle embarqué notre technique d’analyse améliore les estimations WCRT de 21% en moyenne.
En savoir plus

21 En savoir plus

Scheduling in a random environment: stability and asymptotic optimality

Scheduling in a random environment: stability and asymptotic optimality

that have been completed while receiving service for a total duration of time t when being in state n. In order to derive stability and asymptotic optimality results, we will be interested in limits of the fluid-scaled process. In Section IV-A we will characterize a generic description of weak fluid limits (usually not unique) of Equation (3), following the same reasoning as in [15]. In Section IV-B we focus instead on a special class of policies for which we can prove convergence in probability towards a unique limit, which will be referred to as the strong fluid limit. (In [20] similar is done but only for a subset of the state space.) We discuss the differences between weak and strong fluid limits in more detail in Remark 5, after having introduced formally both concepts.
En savoir plus

15 En savoir plus

Preemptive scheduling with variable profile, precedence constraints and due dates

Preemptive scheduling with variable profile, precedence constraints and due dates

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

32 En savoir plus

Dynamic Scheduling of Real-Time Tasks on Multicore Architectures

Dynamic Scheduling of Real-Time Tasks on Multicore Architectures

IV. C ONCLUSION AND F UTURE W ORK Our goal of our research is to propose real-time energy-efficient scheduling for embedded many-core architecture with more general recurring task model which can be more appropriate for industrial purpose and to take into account the new trends, for example using processor/core groups. The algorithm presented here aims to minimize preemptions in shared memory systems. We are also working on dynamic scheduling with hierarchical memory and allowing both local and global reconfiguration. We would like to define how to evaluate costs for both of them. Another way is to work on thread parallelism: it is a part of a job J i executed simultaneously on several cores/processors
En savoir plus

3 En savoir plus

Test of preemptive real-time systems

Test of preemptive real-time systems

chooseAction and (3) its timing by the function chooseDelay (see Algorithm 1). The rswtioco relation does not allow either of “standard” outputs and ‘’indicators’’ to be emitted in advance or on late, by the system. Also, this relation allows having more information about the non-conformance of a system. So, when the system emits an indicator or an output that was not expected at that time, then we can know if that indicator (resp. output) must be an active output (resp. an indicator) or nothing (see algorithm 1). The proposed rswtioco relation makes it possible to answer another question: “does some action a resume at the expected date? i.e. rswtioco does not allow a suspended action to be resumed in advance or on late. Under assumptions of input enabledness, the rswtioco relation coincides with relativized timed trace inclusion. Timed Traces of the SUT operating under an environment must be included in those of the specification under the cover of the same environment.
En savoir plus

22 En savoir plus

Show all 10000 documents...