## Probabilistic Analysis and Scheduling of Critical Soft Real-Time Systems (1999)

Citations: | 22 - 0 self |

### BibTeX

@MISC{Gardner99probabilisticanalysis,

author = {Mark K. Gardner},

title = { Probabilistic Analysis and Scheduling of Critical Soft Real-Time Systems},

year = {1999}

}

### OpenURL

### Abstract

In addition to correctness requirements, a real-time system must also meet its temporal constraints, often expressed as deadlines. We call safety or mission critical real-time systems which may miss some deadlines critical soft real-time systems to distinguish them from hard real-time systems, where all deadlines must be met, and from soft real-time systems which are not safety or mission critical. The performance of a critical soft real-time system is acceptable as long as the deadline miss rate is below an application specific threshold. Architectural features of computer systems, such as caches and branch prediction hardware, are designed to improve average performance. Deterministic real-time design and analysis approaches require that such features be disabled to increase predictability. Alternatively, allowances must be made for for their effects by designing for the worst case. Either approach leads to a decrease in average performance. Since critical soft real-time systems do not require that all deadlines be met, average performance can be improved by adopting a probabilitistic approach. In order to allow a trade-off between deadlines met and average

### Citations

9158 | Introduction to Algorithms
- Cormen, Leiserson, et al.
- 1998
(Show Context)
Citation Context ...�, where��and À�are the Fourier transforms of�Øand�Ørespectively. The result is an ÇÆÐÓ�Æalgorithm for convolution. There are many descriptions and implementations of the FFT readily available (e.g., =-=[51, 79, 80]-=-). Three issues need to be considered when using FFT to perform convolution. First, the discrete representations of the probability density functions being convolved must have the same sampling rate a... |

3268 | Scheduling algorithms for multiprogramming in a hard-real-time environment
- Liu, Layland
- 1973
(Show Context)
Citation Context ...ctives and Contributions The objective of this thesis is to develop techniques for analyzing and scheduling critical soft real-time systems. Specifically, this work 1. extends the periodic task model =-=[1]-=- to describe systems of hard, soft and non-real-time tasks in a uniform manner, 2. develops an analysis method for bounding the frequency of missed deadlines for three classes of fixed priority system... |

1824 | A generalized processor sharing approach to flow control—the single node case - Parekh, Gallager - 1993 |

1722 |
others, The Java Language Specification
- Gosling
(Show Context)
Citation Context ...er specifying its handler would reduce the amount of code that needs to be written to create a simulation. It is recommended that programs written in Java should make extensive use of inner classes ( =-=[72]-=-) to specialize behavior rather than use inheritance. 4 Even with inner classes, it is still not as convenient to parameterize behavior in Java as one would like. Inner classes are a heavy weight solu... |

1160 | Analysis and simulation of a fair queueing algorithm
- Demers, Keshav, et al.
- 1989
(Show Context)
Citation Context ...for fixed priority scheduling and 15sConstant Utilization Servers [35], Total Bandwidth Servers [21] or Weighted Fair Queueing Servers [39] (also called Packet-by-Packet Generalized Processor Sharing =-=[40]-=-) for dynamic priority scheduling. They observe that the average response time of an aperiodic task decreases with increases in Sporadic Server server period while we observe that increasing the serve... |

916 | Priority Inheritance Protocols: An Approach to Real-Time Synchronization - Sha, Rajkumar, et al. - 1990 |

645 |
The rate monotonic scheduling algorithm: Exact characterization and average case behavior
- Lehoczky, Sha, et al.
- 1989
(Show Context)
Citation Context ... of the task and higher priority tasks is greater than the multiframe bound. In addition, some systems do not have a fixed pattern of worst-case execution times. The Time Demand Analysis (TDA) method =-=[9]-=- provides a more accurate and 9sgeneral characterization of the schedulability of arbitrary fixed-priority systems than the Liu and Layland bound. For a system of independent periodic tasks in which t... |

585 | Virtual Clock: A New Traffic Control Algorithm for Packet Switching Networks - Zhang - 1990 |

463 |
Empirical processes with applications to statistics
- Shorack, Wellner
- 1986
(Show Context)
Citation Context ...rovide a bound on the probability distribution function of a random variable when only the mean or standard deviation is known. For a derivation of the inequality or for alternative formulations, see =-=[46, 74, 75]-=-. È��� ���Æ℄��Æ Lemma D.1.1 Given a random variable�with mean�and variance�,the probability that a value of�differs from the mean by more thanÆ�is D.2 Bennett Inequality The Bennett Inequality, as sta... |

380 |
On the complexity of fixed-priority scheduling of periodic,real-time tasks
- Leung, Whitehead
- 1982
(Show Context)
Citation Context ...rrange the tasks in order of non-increasing priorityÌ��Ì� such thatÌ� has a higher priority thanÌ� for all�. Examples of fixed priority policies are Rate Monotonic (RM) [1] or Deadline Monotonic (DM) =-=[30]-=-. The priority of a task under RM is inversely proportional to the period of the task. The priority of a 7stask under DM is inversely proportional to the relative deadline of the task. Priorities may ... |

367 | Applying New Scheduling Theory To Static Priority Pre-emptive Scheduling - Audsley, Burns, et al. - 1993 |

365 |
Aperiodic task scheduling for hardreal-time systems
- Sprunt, Sha, et al.
- 1989
(Show Context)
Citation Context ...execution time in our model, and a request to a server for the execution of the remaining portion. Under fixed priority scheduling, OSM and TTM both execute the remaining portion by a Sporadic Server =-=[20]-=-. Under an EDF scheduler, OSM executes requests by either a Constant Utilization Server [35] or a Total Bandwidth Server [21] rather than a Slack Stealer [11] as is the case with TTM. The Overload Ser... |

328 |
Fundamental Design Problems of Distributed Systems for the Hard Real-Time Environment
- Mok
- 1983
(Show Context)
Citation Context ... the Stochastic Time Demand Analysis method so that it can deal with mutual exclusion and focus on systems where all resource accesses are made according to the Non-Preemptable Section (NPS) protocol =-=[13]-=-. According to the NPS protocol, a job in the system cannot be preempted while 43saccessing shared resources, i.e., while it is in a critical section. Jobs with a higher priority released after the cu... |

264 | A Practitioner's Handbook for Real-Time Analysis – Guide to Rate Monotonic Analysis for RealTime Systems - Klein, Ralya, et al. - 1993 |

260 |
Fixed Priority Scheduling of Periodic Task Sets with Arbitrary Deadlines
- Lehoczky
- 1990
(Show Context)
Citation Context ...or execution. It ends at the first time instantØat which all jobs inÌ�and higher priority tasks released beforeØhave completed. Analogous to the critical instant analysis in [1], it has been shown in =-=[10]-=- that it suffices for us to consider only an in-phase level-��busy interval for the following reasons. 1. If a job inÌ�is ever released at the same time as a job in every higher priority task, that in... |

242 | Algorithms for scheduling imprecise computations - Liu, Lin, et al. - 1991 |

207 | C.: A proportional share resource allocation algorithm for real-time, time-shared systems
- Stoica, Abdel-Wahab, et al.
- 1996
(Show Context)
Citation Context ...the average response times as our metrics instead of average error. The Isolation Server Method (ISM), proposed in Chapter 5, is similar to the Proportional Share Resource Allocation algorithm (PSRA) =-=[37]-=-. Both assign a portion of the processor bandwidth to a task. Whereas the PSRA algorithm allocates the assigned portion to jobs in discrete-sized time quanta, the ISM allocates the portion in variable... |

193 |
An Optimal Algorithm for Scheduling Soft-aperiodic Tasks
- Lehoczky, Ramos-Thuel
- 1992
(Show Context)
Citation Context ...he remaining portion by a Sporadic Server [20]. Under an EDF scheduler, OSM executes requests by either a Constant Utilization Server [35] or a Total Bandwidth Server [21] rather than a Slack Stealer =-=[11]-=- as is the case with TTM. The Overload Server Method is also similar to the work by Chung et al. [36]; the main difference being that the remaining portions of all jobs execute to completion under OSM... |

190 | Latency-rate servers: a general model for analysis of traffic scheduling algorithms
- Stiliadis, Varma
- 1998
(Show Context)
Citation Context ...vation by the same arbitrary amount. This result is not surprising since TBS is a preemptable variant of the Virtual Clock (VC) network packet scheduling algorithm [61–63] which is known to be unfair =-=[59, 64]-=-. Note that fairness is a more stringent requirement on servers than ensuring that tasks receive their guaranteed utilizations on the average. It implies that tasks are allocated an amount of time at ... |

190 | Comparison of rate-based service disciplines
- Zhang, Keshav
- 1991
(Show Context)
Citation Context ...ads with significant release time jitter. (See [60, 64, 65] for further discussions of the properties of the latency-rate and rate-proportional classes of servers to which WFQ and VC belong. See also =-=[66]-=- for a comparison of other network scheduling algorithms.) We now describe WFQ servers by first describing the Generalized Processor Sharing (GPS) algorithm [59,67,68] that they approximate. The GPS a... |

183 | Processor Capacity reserves: Operating System Support for Multimedia Applications
- Mercer, Savage, et al.
- 1994
(Show Context)
Citation Context ... number of deadlines met and minimize response time. We note that specifying an execution time for a task less than the maximum is not new. Under the Processor Capacity Reserve model of Mercer et al. =-=[52]-=-, a task is guaranteed to execute for at least its guaranteed execution time each period. However, we appear to be the first to use the specification of a guaranteed execution time equal to or less th... |

158 |
Rao,Normal Approximation and Asymptotic Expansions
- Bhattacharya, Ranga
- 1976
(Show Context)
Citation Context ...�� � of�Ò, ¡¡¡�Ò, by more thanÆ� whereÆ��ÔÒÚÒ�Ú È��Ò �Ò�Æ℄��ÜÔ� Ú ÚÒsÆÆ� is ¡¡¡ÚÒ. Note that���is suggested in [75], if��� �����. D.4 Berry-Esseen Inequality The Berry-Esseen Inequality, as stated in =-=[75, 76]-=-, requires a zero mean for each random variable in a sum of independent random variables. The formulation given here allows a non-zero mean and follows by way of algebraic manipulation. 183sof�Ò�� � ¡... |

141 |
Fast Fourier Transformation and Convolution Algorithms, 2nd ed
- Nussbaumer
- 1982
(Show Context)
Citation Context ...he general case, the probability density function of the sum must be computed by convolution. Convolution, however, has remained computationally expensive in spite of efforts to reduce its cost. (See =-=[51]-=-, for example). The fastest known method for performing convolution is to compute the Fast Fourier Transforms of the density functions, multiply, and and compute the inverse transform. This is still a... |

139 |
Real-Time Synchronization Protocol for Multiprocessors
- Sha, Rajkumar, et al.
- 1998
(Show Context)
Citation Context ...has entered a critical section are blocked until the currently executing job exits the critical section. A priority inversion is said to occur whenever a job waits while a lower priority job executes =-=[14]-=-. Uncontrolled priority inversion cannot occur under the NPS protocol if the duration of every critical section is bounded. This fact is formally stated by the following lemma, the proof of which can ... |

138 | Synchronization in Real-Time Systems: A Priority Inheritance Approach - Rajkumar - 1991 |

126 |
Scheduling periodic jobs that allow imprecise results
- Chung, Liu, et al.
- 1990
(Show Context)
Citation Context ...er a Constant Utilization Server [35] or a Total Bandwidth Server [21] rather than a Slack Stealer [11] as is the case with TTM. The Overload Server Method is also similar to the work by Chung et al. =-=[36]-=-; the main difference being that the remaining portions of all jobs execute to completion under OSM instead of being terminated when deadlines cannot be met. Chung also investigated several policies f... |

126 | A new architecture for packet switching network protocols - Zhang - 1989 |

124 | A multiframe model for real-time tasks
- Mok, Chen
- 1997
(Show Context)
Citation Context ...and Layland bound is that it assumes the worst-case execution time for every job in a task and hence may be overly pessimistic when execution times vary widely. In their paper on the Multiframe Model =-=[31]-=-, Mok and Chien observed that the execution times of jobs in many tasks vary according to a fixed repeating pattern. For example, a task which decodes a MPEG video stream executes longer to decodeÁfra... |

102 |
Probabilistic Performance Guarantee for RealTime Tasks with Varying Computation Times
- Tia, Deng, et al.
- 1995
(Show Context)
Citation Context ...nalysis There are only two other real-time techniques that exploit information about the statistical behavior of periodic tasks to analyze real-time systems: Probabilistic Time Demand Analysis (PTDA) =-=[32]-=- and Statistical Rate Monotonic Scheduling (SRMS) [33]. Like the method proposed in Chapter 4, PTDA attempts to provide a lower bound on the probability that jobs in a task will complete in time. It i... |

101 | Elastic task model for adaptive rate control - Buttazzo, Lipari, et al. - 1998 |

101 | Deadline Assignment in a Distributed Soft Real-Time System - Kao, Molina - 1997 |

91 |
Congestion Control in Computer Networks
- Keshav
- 1991
(Show Context)
Citation Context ...to a dynamic priority environment, while we use Sporadic Servers for fixed priority scheduling and 15sConstant Utilization Servers [35], Total Bandwidth Servers [21] or Weighted Fair Queueing Servers =-=[39]-=- (also called Packet-by-Packet Generalized Processor Sharing [40]) for dynamic priority scheduling. They observe that the average response time of an aperiodic task decreases with increases in Sporadi... |

88 |
Aperiodic Servers In A Deadline Scheduling Environment,” Real-Time Sys- tems
- Ghazalie, Baker
- 1995
(Show Context)
Citation Context ... Chapter 5, our results indicate that the behavior of a system with multiple servers is bounded by the assignment of all tasks to a single server and the assignment of each task to its own server. In =-=[6]-=-, Ghazalie and Baker consider the performance of several aperiodic servers in a deadline-driven environment. Their focus is on scheduling aperiodic tasks while our focus is on scheduling overruns usin... |

86 | Statistical Rate Monotonic Scheduling
- Atlas, Bestavros
- 1998
(Show Context)
Citation Context ...that exploit information about the statistical behavior of periodic tasks to analyze real-time systems: Probabilistic Time Demand Analysis (PTDA) [32] and Statistical Rate Monotonic Scheduling (SRMS) =-=[33]-=-. Like the method proposed in Chapter 4, PTDA attempts to provide a lower bound on the probability that jobs in a task will complete in time. It is a straight forward extension to TDA in which the tim... |

85 | CPU service classes for multimedia applications
- Chu, Nahrstedt
- 1999
(Show Context)
Citation Context ...obs must be executed and they must complete in order. We allow either fixed or dynamic priority assignments. 16sThe work by Chu and Nahrstedt on the Dynamic Soft Real-Time scheduling framework (DSRT) =-=[43, 44]-=- is also closely related. Their middleware, which requires conformance to the POSIX 1003.1b standard for real-time support, requires no modifications to the kernel in order to co-schedule hard, soft, ... |

80 | On satisfying timing constraints in hard real-time systems - Xu, Parnas - 1993 |

78 | An extendible approach for analyzing fixed priority hard real-time tasks - Tindell, Burns, et al. - 1994 |

76 | A stack-based resource allocation policy for realtime processes - Baker - 1990 |

69 |
Queueing Systems, Volume I
- Kleinrock
- 1975
(Show Context)
Citation Context ...s the “memoryless property” of the arrival time and/or execution time distributions. For example, M/M/1 and M/G/1 queues have Poisson arrivals; M/M/1 and G/M/1 queues have exponential execution times =-=[46, 47]-=-. The memoryless property makes the analysis of otherwise complex queueing systems tractable. On the other hand, jobs of a real-time task arrive more or less periodically, implying an arbitrary inter-... |

69 |
Real-Time Queueing Theory
- Lehoczky
- 1996
(Show Context)
Citation Context ... is based requires the average number of jobs in the queue to be large in order to be accurate. (The results for a 95% average system utilization and uniformly distributed execution times reported in =-=[48]-=- were obtained with an average queue length of 50 jobs. In contrast, the system 19sof three tasks scheduled by a Total Bandwidth Server discussed in Figure 5.8 of Chapter 5 has a maximum queue length ... |

68 | Rate-proportional servers: a design methodology for fair queueing algorithms
- Stiliadis, Varma
(Show Context)
Citation Context ...rage than an unfair scheduler. A scheduler is said to be fair when the normalized service received by each task which is backlogged throughout an interval differs by no more than a fairness threshold =-=[60]-=-,��. (A task is backlogged throughout an interval if at any time in the interval at least one job from the task is eligible for execution.) denote the service (i.e., amount of processor time) received... |

67 | Real-time synchronization protocols for shared memory multiprocessors - Rajkumar - 1990 |

63 | Algoithms for Scheduling Hard Aperiodic Tasks in fixed-Priority Systems Using Slack Stealing,” Real-Time Systems Symposium - Ramos-Thuel, Lehoczky - 1994 |

63 | Overload management in real-time control applications using (m,k)-firm guarantee
- Ramanathan
- 1999
(Show Context)
Citation Context ...ether on a fast processor without causing missed deadlines. The primary motivation of the OSM and the ISM is to accommodate overrun. Thus, the OSM and the ISM are complimentary to the OSE. Ramanathan =-=[41]-=- reduces the load on an overloaded system by selectively discarding jobs in a task according to theÑ��-firm deadline model of [42]. According to the approach, the stream of jobs from a task is partiti... |

63 | Allocating real-time tasks: An NP-hard problem made easy - Tindell, Burns, et al. - 1992 |

59 | A scheme for scheduling hard real-time application in open system environment
- Deng, Liu, et al.
- 1997
(Show Context)
Citation Context ...rtion. Under fixed priority scheduling, OSM and TTM both execute the remaining portion by a Sporadic Server [20]. Under an EDF scheduler, OSM executes requests by either a Constant Utilization Server =-=[35]-=- or a Total Bandwidth Server [21] rather than a Slack Stealer [11] as is the case with TTM. The Overload Server Method is also similar to the work by Chung et al. [36]; the main difference being that ... |

59 | PERTS: A Prototyping Environment for Real-Time Systems
- Liu, Liu, et al.
- 1996
(Show Context)
Citation Context ...d, but If each the probability that a from the mean of�Ò, denoted 184sAppendix E Implementing STDA In this chapter, we discuss an implementation of STDA in the PERTS real-time prototyping environment =-=[77, 78]-=-. PERTS is a tool which facilitates the design and analysis of real-time systems by applying theoretical results, where possible, or by simulating the system to determine its behavior. The issues we d... |

58 | Efficient fair queueing algorithms for packetswitched networks
- Stiliadis, Varma
- 1998
(Show Context)
Citation Context ...o schedule network packets [39,40]. Because of improved fairness, Weighted Fair Queueing Servers (WFQS) are expected to perform better than TBS on workloads with significant release time jitter. (See =-=[60, 64, 65]-=- for further discussions of the properties of the latency-rate and rate-proportional classes of servers to which WFQ and VC belong. See also [66] for a comparison of other network scheduling algorithm... |

54 | Robust Aperiodic Scheduling Under Dynamic Priority Systems
- Spuri, Buttazzo, et al.
- 1995
(Show Context)
Citation Context ...uling, OSM and TTM both execute the remaining portion by a Sporadic Server [20]. Under an EDF scheduler, OSM executes requests by either a Constant Utilization Server [35] or a Total Bandwidth Server =-=[21]-=- rather than a Slack Stealer [11] as is the case with TTM. The Overload Server Method is also similar to the work by Chung et al. [36]; the main difference being that the remaining portions of all job... |

52 |
Queueing Systems, Volume II
- Kleinrock
- 1976
(Show Context)
Citation Context ...s the “memoryless property” of the arrival time and/or execution time distributions. For example, M/M/1 and M/G/1 queues have Poisson arrivals; M/M/1 and G/M/1 queues have exponential execution times =-=[46, 47]-=-. The memoryless property makes the analysis of otherwise complex queueing systems tractable. On the other hand, jobs of a real-time task arrive more or less periodically, implying an arbitrary inter-... |