Utility maximizing dynamic resource management in an oversubscribed energy-constrained heterogeneous computing system

Khemka, Bhavesh; Friese, Ryan; Pasricha, Sudeep; Maciejewski, Anthony A.; Siegel, Howard Jay; Koenig, Gregory A.; Powers, Sarah; Hilton, Marcia; Rambharos, Rajendra; Poole, Steve

doi:10.1016/j.suscom.2014.08.001

Cited by 33 publications

(53 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further, energy can also be optimized by performing energy-aware resource allocation [23]. The approaches considering DVFS and optimizing both the value and energy consumption are recently reported [5], [11], but they do not perform adaptive resource allocation as job execution status and available system cores are not monitored.…”

Section: Related Workmentioning

confidence: 99%

“…Each job is considered to have a soft real-time deadline, implying that the violation of deadline does not make the computation irrelevant, but reduces its value for the user [5], [6], [8]. The value of the job to the user depends upon the completion time and is represented by value curve.…”

Section: A Job and Its Value Curve Modelmentioning

confidence: 99%

“…remainedT ime = remainedW ork × completionT ime usedCores (3) newEndT ime = progreesedT ime + remainedT ime (4) newV alueDensity newEnergyDensity = newV alue/remainedT ime newEnergy/remainedT ime (5) where completionT ime usedCores is the completion time at the used number of cores and remainedW ork is computed by looking the deviation included end time and the deviation at the current checkpoint, and it is normalized with respect to the total work to be done from start to end time of the job. The progreesedT ime is the job progressed time by the checkpoint.…”

Section: ) Adaptation (Lines 7-17)mentioning

confidence: 99%

“…A job may contain a number of dependent/independent tasks or processes. The notion of values (economic or otherwise) of jobs has been introduced to define their importance level [5]. In overload situations where demand for available resources is higher than the supply, such a notion facilitates in deciding to allocate limited resources to the high value jobs and holding the low value jobs for late allocation.…”

Section: Introductionmentioning

confidence: 99%

“…If the job is completed before the deadline, a high value is achieved; otherwise, a low value is achieved. This also implies that the violation of deadline does not make the computation irrelevant, but reduces its value for the user [5], [8], [9]. Deadlines missed by large margins may result in zero value and thus the computation becomes useless for the user.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Value and Energy Aware Adaptive Resource Allocation of Soft Real-Time Jobs on Many-Core HPC Data Centers

Singh

Dziurzański

Indrusiak

2016

2016 IEEE 19th International Symposium on Real-Time Distributed Computing (ISORC)

View full text Add to dashboard Cite

Abstract-Modern high performance computing (HPC) data centers consume huge energy to operate them. Therefore, appropriate measures are required to reduce their energy consumption. Existing efforts for such measures focus on consolidation and dynamic voltage and frequency scaling (DVFS). However, most of them do not perform adaptive resource allocation for the executing dependent tasks (or jobs) in order to optimize both value and energy. The value is achieved by completing the execution of a job and it depends on the completion time. A high value is achieved if the job is completed before its deadline; otherwise a lower value. In this paper, we propose an adaptive resource allocation approach that uses design-time profiling results of jobs for efficient allocation and adaptation in order to optimize both value and energy while executing dependent tasks. The profiling results for each job are obtained by exploiting efficient allocation combined with identification of voltage/frequency levels of used system cores and used in adapting to different number of cores based on the monitored execution progress of the job and available cores. Experiments show that the proposed approach enhances the overall value by about 10% when compare to existing approaches while showing reduction in energy consumption and percentage of rejected jobs leading to zero value.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: A Job and Its Value Curve Modelmentioning

confidence: 99%

Section: ) Adaptation (Lines 7-17)mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Value and Energy Aware Adaptive Resource Allocation of Soft Real-Time Jobs on Many-Core HPC Data Centers

Singh

Dziurzański

Indrusiak

2016

2016 IEEE 19th International Symposium on Real-Time Distributed Computing (ISORC)

View full text Add to dashboard Cite

show abstract

Controlling the correlation of cost matrices to assess scheduling algorithm performance on heterogeneous platforms

Canon

Héam

Philippe

2017

Concurrency and Computation

View full text Add to dashboard Cite

Summary Bias in the performance evaluation of scheduling heuristics has been shown to undermine the scope of existing studies. Improving the assessment step leads to stronger scientific claims when validating new optimization strategies. This article considers the problem of allocating independent tasks to unrelated machines such as to minimize the maximum completion time. Testing heuristics for this problem requires the generation of cost matrices that specify the execution time of each task on each machine. Numerous studies showed that the task and machine heterogeneities belong to the properties impacting heuristics performance the most. This study focuses on orthogonal properties, the average correlations between each pair of rows and each pair of columns, which measure the proximity with uniform instances. Cost matrices generated with 2 distinct novel generation methods show the effect of these correlations on the performance of several heuristics from the literature. In particular, EFT performance depends on whether the tasks are more correlated than the machines and HLPT performs the best when both correlations are close to one.

show abstract

Communication-Aware Prediction-Based Online Scheduling in High-Performance Real-Time Embedded Systems

Goupille-Lescar

Lenormand

Parlavantzas

et al. 2018

Algorithms and Architectures for Parallel Processing

View full text Add to dashboard Cite

Current high-end, data-intensive real-time embedded sensor applications (e.g., radar, optronics) require very specific computing platforms. The nature of such applications and the environment in which they are deployed impose numerous constraints, including realtime constraints, and computing throughput and latency needs. Static application placement is traditionally used to deal with these constraints. However, this approach fails to provide adaptation capabilities in an environment in constant evolution. Through the study of an industrial radar use-case, our work aims at mitigating the aforementioned limitations by proposing a low-latency online resource manager derived from techniques used in large-scale systems, such as cloud and grid environments. The resource manager introduced in this paper is able to dynamically allocate resources to fulfill requests coming from several sensors, making the most of the computing platform while providing guaranties on non-functional properties and Quality of Service (QoS) levels. Thanks to the load prediction implemented in the manager, we are able to achieve a 83% load increase before overloading the platform while managing to reduce ten times the incurred QoS penalty. Further methods to reduce the impact of the overload are as well as possible future improvements are proposed and discussed.

show abstract

Utility maximizing dynamic resource management in an oversubscribed energy-constrained heterogeneous computing system

Cited by 33 publications

References 14 publications

Value and Energy Aware Adaptive Resource Allocation of Soft Real-Time Jobs on Many-Core HPC Data Centers

Value and Energy Aware Adaptive Resource Allocation of Soft Real-Time Jobs on Many-Core HPC Data Centers

Controlling the correlation of cost matrices to assess scheduling algorithm performance on heterogeneous platforms

Communication-Aware Prediction-Based Online Scheduling in High-Performance Real-Time Embedded Systems

Contact Info

Product

Resources

About