Resource Demand Prediction of Cloud Workloads Using an Attention-based GRU Model

Shu, Wenjuan; Zeng, Fanxin; Ling, Zhen; Liu, Junyi; Lü, Tingting; Chen, Guozhu

doi:10.1109/msn53354.2021.00071

Cited by 9 publications

(6 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Efficiently managing cloud resources relies heavily on accurately predicting resource demands for online tasks, a practice essential for effective resource allocation and task scheduling [29]. Workload prediction is extensively employed to enhance resource management in cloud environments [28,31,35] with comprehensive reviews provided by [32] and [2].…”

Section: Related Workmentioning

confidence: 99%

Forecasting GPU Type Allocation via Demand Feature Extraction in Heterogeneous Clusters

Wang,

Shi,

Chen

et al. 2024

Preprint

View full text Add to dashboard Cite

The availability of multiple types of GPUs enhances parallel processing capabilities and provides users with a broader spectrum of options in heterogeneous GPU clusters. However, the inherent randomness and uncertainty in GPU type requests introduce complexity into resource allocation, resulting in disparities between requested and allocated GPU types and exacerbating imbalances in the GPU resource distribution. In this study, we perform a comprehensive analysis of the attributes associated with diverse GPU requests and allocations in heterogeneous clusters. By introducing the adaptive selection model (ASM), our approach not only predicts the demand for various GPU types but also extracts critical features from the resource demand. These features play a pivotal role in predicting GPU resource allocation, effectively bridging the gap between request and allocation. Moreover, we introduce spatial and temporal long short-term memory (ST-LSTM), a novel LSTM-based model that accommodates both spatial and temporal correlations in GPU type allocation. Our experimental efforts confirm the effectiveness of our proposed methodology through the utilization of real-world trace data from Alibaba cloud data centres. The ASM achieves an impressive accuracy rate of 87% and illuminates the key factors influencing user GPU type selection. Additionally, the ST-LSTM model demonstrates excellent performance with average RMSE and MAE values of 1.84 and 1.09, respectively.

show abstract

Section: Related Workmentioning

confidence: 99%

Forecasting GPU Type Allocation via Demand Feature Extraction in Heterogeneous Clusters

Wang,

Shi,

Chen

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…It will affect the pricing strategy of the UEs in the game. It is noted that the fluctuation of the task load in the UE over a short period (e.g., a few tens of seconds) has some kind of randomness, but the task loads in the adjacent time periods may have a strong correlation [12]. Next, we propose an FSDT-MC based method [9] to predict the dynamic of f ex k,t during the FL period.…”

Section: A Energy Cost For Model Trainingmentioning

confidence: 99%

“…1) From ref. [12], we know that the workload fluctuation of a machine over a longer period (e.g., a few minutes) has a strong periodicity, while the workload fluctuation over a shorter period (e.g., a few tens of seconds) has a certain randomness, but the workloads in the adjacent time periods have a strong correlation. 2) In this paper, the time spent for a single round of local training is T trn = 2s, the time spent for a single round of parameter transmission is T com = 0.2s, and the total number of local training sessions is I g = 10.…”

Section: A Energy Cost For Model Trainingmentioning

confidence: 99%

Task-Load-Aware Game-Theoretic Framework for Wireless Federated Learning

et al. 2023

View full text Add to dashboard Cite

Federated learning (FL) can protect data privacy but has difficulties in motivating user equipment (UE) to engage in task training. This paper proposes a Bertrand-game based framework to address the incentive problem, where a model owner (MO) issues an FL task and the employed UEs help train the model by using their local data. Specially, we consider the impact of time-varying task load and channel quality on UE's motivation to engage in the FL task. We adopt the finitestate discrete-time Markov chain (FSDT-MC) to predict these parameters during the FL task. Depending on the performance metrics set by the MO and the estimated energy cost of the FL task, each UE seeks to maximize its profit. We obtain the Nash equilibrium (NE) of the game in closed form, and develop a distributed iterative algorithm to find it. Finally, simulation result verifies the effectiveness of the proposed approach.

show abstract

“…were used for load forecasting. However, due to its shallow structure, the essential characteristics of the data set cannot be obtained, so the prediction accuracy faces a bottleneck [7] . In recent years, deep learning has gradually become the mainstream method in resource and load prediction [8] .…”

Section: Introductionmentioning

confidence: 99%

Research on cloud-based safe computing platform resource prediction

Hao,

2024

Seventh International Conference on Traffic Engineering and Transportation System (ICTETS 2023)

View full text Add to dashboard Cite

In order to ensure the safe operation of cloud-based safe computing platform in train control system, it is crucial to accurately predict cloud platform resources. Due to the different number of trains at different times and sections, the signal software load in cloud based security computing platforms also varies. Resource data often accompanies high randomness and non-stationary characteristics, increasing the difficulty of prediction. A hybrid prediction model based on Sparse Search Algorithm (SSA), Variational Mode Decomposition (VMD), SE module, and Time Convolutional Network (TCN) is proposed to address the low prediction accuracy of traditional prediction methods and the lack of research on cloud platform in train control system. First, the original data is decomposed using the VMD optimized by SSA, and then each component is predicted using the SE-TCN model. Finally, integrate the prediction results of each component to obtain the final cloud resource usage prediction result. The experimental results show that, compared with other models, the Mean absolute error (MAE) of the proposed model on the memory data are reduced by 27%, and the CPU data are reduced by 24%, with high prediction accuracy.

show abstract

Resource Demand Prediction of Cloud Workloads Using an Attention-based GRU Model

Cited by 9 publications

References 17 publications

Forecasting GPU Type Allocation via Demand Feature Extraction in Heterogeneous Clusters

Forecasting GPU Type Allocation via Demand Feature Extraction in Heterogeneous Clusters

Task-Load-Aware Game-Theoretic Framework for Wireless Federated Learning

Research on cloud-based safe computing platform resource prediction

Contact Info

Product

Resources

About