Algorithms for scheduling imprecise computations

Liu, J.W.S.; Lin, Kwei-Jay; Shih, Wei-Kuan; Yu, Albert; Chung, Jen-Yao; Zhao, Wei

doi:10.1109/2.76287

Cited by 386 publications

(131 citation statements)

References 10 publications

Supporting

Mentioning

124

Contrasting

Unclassified

Order By: Relevance

“…For instance, this happens with monotonic queries and transactions, which are the counterparts of monotonic tasks [34] in real-time systems. Furthermore, many of the extensions to serializability that have been proposed in databases are also applicable to real-time databases (see [38] for a review of these proposals).…”

Section: Databases and Real-time Systemsmentioning

confidence: 98%

Real-time databases

Ramamritham

1993

Distrib Parallel Databases

440

209

View full text Add to dashboard Cite

Abstract. Data in real-time databases has to be logically consistent as well as temporally consistent.The latter arises from the need to preserve the temporal validity of data items that reflect the state of the environment that is being controlled by the system. Some of the timing constraints on the transactions that process real-time data come from this need. These constraints, in turn, necessitate time-cognizant transaction processing so that transactions can be processed to meet their deadlines.This paper explores the issues in real-time database systems and presents an overview of the state of the art. After introducing the characteristics of data and transactions in real-time databases, we discuss issues that relate to the processing of time-constrained transactions. Specifically, we examine different approaches to resolving contention over data and processing resources. We also explore the problems of recovery, managing I/O, and handling overloads. Real-time databases have the potential to trade off the quality of the result of a query or a transaction for its timely processing. Quality can be measured in terms of the completeness, accuracy, currency, and consistency of the results. Several aspects of this trade-off are also considered.

show abstract

Section: Databases and Real-time Systemsmentioning

confidence: 98%

Real-time databases

Ramamritham

1993

Distrib Parallel Databases

440

209

View full text Add to dashboard Cite

show abstract

“…Such knobs may be in the form of sampling rates and iterative improvement functions which affect the quality in terms of metrics such as accuracy, coverage area, certainty, and resolution (level of detail). Quality-time tradeoffs may be achieved by several techniques such as milestone method, sieve functions and selecting between multiple versions of the algorithm [10]. With the milestone method, the algorithm executes for a minimum duration and then evaluates its progress at checkpoints within the controlflow graph.…”

Section: A Anytime Algorithmsmentioning

confidence: 99%

“…Following this, several studies in the AI community focused on composing anytime algorithms to more complex systems for sensor interpretation and path planning [21], [22], search [8], and evaluation of belief networks [23]. In the early 1990s, the real-time community explored scheduling approaches for imprecise and approximate computing [10], [13]. The Flex language was developed with time as a first-class citizen to specify timing and performance requirements within the algorithm [11], [12].…”

Section: Related Workmentioning

confidence: 99%

Anytime Algorithms for GPU Architectures

Mangharam

Saba

2011

2011 IEEE 32nd Real-Time Systems Symposium

View full text Add to dashboard Cite

Most algorithms are run-to-completion and provide one answer upon completion and no answer if interrupted before completion. On the other hand, anytime algorithms have a monotonic increasing utility with the length of execution time. Our investigation focuses on the development of time-bounded anytime algorithms on Graphics Processing Units (GPUs) to trade-off the quality of output with execution time. Given a time-varying workload, the algorithm continually measures its progress and the remaining contract time to decide its execution pathway and select system resources required to maximize the quality of the result. To exploit the quality-time tradeoff, the focus is on the construction, instrumentation, on-line measurement and decision making of algorithms capable of efficiently managing GPU resources. We demonstrate this with a Parallel A* routing algorithm on a CUDA-enabled GPU. The algorithm execution time and resource usage is described in terms of CUDA kernels constructed at design-time. At runtime, the algorithm selects a subset of kernels and composes them to maximize the quality for the remaining contract time. We demonstrate the feedback-control between the GPU-CPU to achieve controllable computation tardiness by throttling request admissions and the processing precision. As a case study, we have implemented AutoMatrix, a GPU-based vehicle traffic simulator for real-time congestion management which scales up to 16 million vehicles on a US street map. This is an early effort to enable imprecise and approximate real-time computation on parallel architectures for stream-based timebounded applications such as traffic congestion prediction and route allocation for large transportation networks. ©2011 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. Abstract-Most algorithms are run-to-completion and provide one answer upon completion and no answer if interrupted before completion. On the other hand, anytime algorithms have a monotonic increasing utility with the length of execution time. Our investigation focuses on the development of time-bounded anytime algorithms on Graphics Processing Units (GPUs) to trade-off the quality of output with execution time. Given a time-varying workload, the algorithm continually measures its progress and the remaining contract time to decide its execution pathway and select system resources required to maximize the quality of the result. To exploit the quality-time tradeoff, the focus is on the construction, instrumentation, on-line measurement and decision making of algorithms capable of efficiently managing GPU resources. We demonstrate this with a Parallel A* routing algorithm on a CUDA-enabled GPU. The algorithm execution time and resource usage is described in terms of CUDA kernels cons...

show abstract

“…The reason is that the general problem of choosing updates and considering data relationships is NP-hard in the strong sense (see section 2.3) and previous on-line algorithms are simplified to reduce computational complexity in such a way that they reject updates when the updates cannot be fitted within the available time. Let us assume d 9 in figure 1 is about Tasks in the imprecise computation model can be described with one of the following approaches [22]. Milestone approach: The result of a task is refined as its execution progresses.…”

Section: Related Workmentioning

confidence: 99%

Data Freshness and Overload Handling in Embedded Systems

Gustafsson

Hansson

2006

12th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA'06)

View full text Add to dashboard Cite

In this paper we consider data freshness and overload handling in embedded systems. The requirements on data management and overload handling are derived from an engine control software. Data items need to be up-to-date, and to achieve this data dependencies must be considered, i.e., updating a data item requires other data items are upto-date. We also note that a correct result of a calculation can in some cases be calculated using a subset of the inputs. Hence, data dependencies can be divided into required and not required data items, e.g., only a subset of data items affecting the fuel calculation in an engine control needs to be calculated during a transient overload in order to reduce the number of calculations. Required data items must always be up-to-date, whereas not required data items can be stale. We describe an algorithm that dynamically determines which data items need to be updated taking workload, data freshness, and data relationships into consideration. Performance results show that the algorithm suppresses transient overloads better than (m, k)-and skipover scheduling combined with established algorithms to update data items. The performance results are collected from an implementation of a real-time database on the realtime operating system µC/OS-II. To investigate whether the system is occasionally overloaded an offline analysis algorithm estimating period times of updates is presented.

show abstract

Algorithms for scheduling imprecise computations

Cited by 386 publications

References 10 publications

Real-time databases

Real-time databases

Anytime Algorithms for GPU Architectures

Data Freshness and Overload Handling in Embedded Systems

Contact Info

Product

Resources

About