A hybrid Markov chain model for workload on parallel computers

Krampe, Anne; Lepping, Joachim; Sieben, Wiebke

doi:10.1145/1851476.1851563

Cited by 6 publications

(4 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The algorithm was tested for generating, executing and measuring 10 workloads of 1000 job requests selected from the previously characterized jobs. In order to create the workload, a Markov chain model was used [28] [29]. Each job requires 1 GPU and it is assumed that there is no data dependence between any jobs.…”

Section: Methodsmentioning

confidence: 99%

Power-Aware Job Scheduling on Heterogeneous Multicore Architectures

Vanzolini

Mucci

Scarselli

et al. 2015

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

This paper presents a power-aware scheduling algorithm based on efficient distribution of the computing workload to the resources on heterogeneous CPU-GPU architectures. The scheduler manages the resources of several computing nodes with a view to reducing the peak power. The algorithm can be used in concert with adjustable power state software services in order to further reduce the computing cost during high demand periods. Although our study relies on GPU workloads, the approach can be extended to other heterogeneous computer architectures. The algorithm has been implemented in a real CPU-GPU heterogeneous system. Experiments prove that the approach presented reduces peak power by 10% compared to a system without any power-aware policy and by up to 24% with respect to the worst case scenario with an execution time increase in the range of 2%. This leads to a reduction in the system and service costs.

show abstract

Section: Methodsmentioning

confidence: 99%

Power-Aware Job Scheduling on Heterogeneous Multicore Architectures

Vanzolini

Mucci

Scarselli

et al. 2015

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

show abstract

“…The premise of this evaluation is as follows: if input workloads are similar, then the performance metrics that evaluate the output of the system that is executing the workloads should be similar as well. An elegant way of doing that is to use scheduling metrics [18], because they can be principal targets for performance optimisation in a system. Two workloads that produce the same waiting times or the same execution times can be considered similar in performance evaluation.…”

Section: Methodsmentioning

confidence: 99%

A similarity measure for time, frequency, and dependencies in large-scale workloads

Lassnig

Fahringer

Garonne

et al. 2011

Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis

View full text Add to dashboard Cite

Performance evaluations of large-scale systems require the use of representative workloads with certifiable similar or dissimilar characteristics. To quantify the similarity of the characteristics, we describe a novel measure comprising two efficient methods that are suitable for large-scale workloads. One method uses the discrete wavelet transform to assess the periodic time and frequency characteristics in the workload. The second method evaluates dependencies in descriptive attributes via association rule learning. Both methods are evaluated to find the limits of their similarity spaces. Additionally, the wavelet method is evaluated against existing similarity methods and tested for noise robustness and random bias. An empirical study using workloads from seven operational large-scale systems evaluates the measure's accuracy. The results show that our measure is highly resistant to noise, well-suited for large-scale workloads, covers 87% of the possible similarity space, and improves accuracy by 24.5% and standard deviation by 10.8% when compared to existing work.

show abstract

“…According to previous studies, the PA of multi-markov chain model, as is illustrated in [14], can reach to 53%. PA of hybrid-markov chain model [15] can reach to 70%. However, the hybrid-markov chain model is based on clustering the users into lots of different groups by complex clustering algorithm.…”

Section: Testmentioning

confidence: 98%

Pre-fetching Webpages on Mobile Social Network: User-Aware Dynamic Markov Chain

Zhao

Hong

2012

2012 8th International Conference on Mobile Ad-Hoc and Sensor Networks (MSN)

View full text Add to dashboard Cite

Different users, different servicing is an important commercial strategy. As VIP users are the main source of revenue, how to provide precise and personalized services for them becomes a hot spot for Mobile Social Network(MSN) providers. Effective pre-fetching of web-pages can improve Quality of Experience (QoE) for MSN users by reducing latency perceived from end-to-end. In this paper, we propose a novel user-aware dynamic Markov chain model to provide personalized pre-fetching for VIP users while guaranteeing the common pre-fetching for ordinary users. It can avoid the weak points generated by applying the former pre-fetching mechanisms to MSN: non-user awareness, low accuracy, high complexity, and repetitive training. Based on real click-stream data of wap.renren.com collected from a main Mobile Telecom Carrier in Chongqing province of China, we evaluate the model.

show abstract

A hybrid Markov chain model for workload on parallel computers

Cited by 6 publications

References 18 publications

Power-Aware Job Scheduling on Heterogeneous Multicore Architectures

Power-Aware Job Scheduling on Heterogeneous Multicore Architectures

A similarity measure for time, frequency, and dependencies in large-scale workloads

Pre-fetching Webpages on Mobile Social Network: User-Aware Dynamic Markov Chain

Contact Info

Product

Resources

About