Efficient utilization of processor and memory resources is essential for sustaining performance and energy efficiency in modern computing infrastructures.While earlier research has emphasized CPU utilization forecasti...Efficient utilization of processor and memory resources is essential for sustaining performance and energy efficiency in modern computing infrastructures.While earlier research has emphasized CPU utilization forecasting,joint prediction of CPU and memory usage under real workload conditions remains underexplored.This study introduces a machine learning–based framework for real-time prediction of CPU and RAM utilization using the Google Cluster Trace 2019 v3 dataset.The framework combines Extreme Gradient Boosting(XGBoost)with a MultiOutputRegressor(MOR)to capture nonlinear interactions across multiple resource dimensions,supported by a leakage-safe imputation strategy that prevents bias frommissing values.Nested cross-validation was employed to ensure rigorous evaluation and reproducibility.Experiments demonstrated that memory usage can be predicted with higher accuracy and stability than processor usage.Residual error analysis revealed balanced error distributions and very low outlier rates,while regime-based evaluations confirmed robustness across both low and high utilization scenarios.Feature ablation consistently highlighted the central role of page cache memory,which significantly affected predictive performance for both CPU and RAM.Comparisons with baseline models such as linear regression and random forest further underscored the superiority of the proposed approach.To assess adaptability,an online prequential learning pipeline was deployed to simulate continuous operation.The system preserved offline accuracy while dynamically adapting to workload changes.It achieved stable performance with extremely low update latencies,confirming feasibility for deployment in environments where responsiveness and scalability are critical.Overall,the findings demonstrate that simultaneous modeling of CPU and RAM utilization enhances forecasting accuracy and provides actionable insights for cache management,workload scheduling,and dynamic resource allocation.By bridging offline evaluation with online adaptability,the proposed framework offers a practical solution for intelligent and sustainable cloud resource management.展开更多
Resistive Random Access Memory(ReRAM)-based neural network accelerators have potential to surpass their digital counterparts in computational efficiency and performance.However,design of these accelerators faces a num...Resistive Random Access Memory(ReRAM)-based neural network accelerators have potential to surpass their digital counterparts in computational efficiency and performance.However,design of these accelerators faces a number of challenges including imperfections of the Re RAM device and a large amount of calculations required to accurately simulate the former.We present XB-SIM,a simulation framework for Re RAM-crossbar-based Convolutional Neural Network(CNN)accelerators.XB-SIM can be flexibly configured to simulate the accelerator’s structure and clock-driven behaviors at the architecture level.This framework also includes an Re RAM-aware Neural Network(NN)training algorithm and a CNN-oriented mapper to train an NN and map it onto the simulated design efficiently.Behavior of the simulator has been verified by the corresponding circuit simulation of a real chip.Furthermore,a batch processing mode of the massive calculations that are required to mimic the behavior of Re RAM-crossbar circuits is proposed to fully apply the computational concurrency of the mapping strategy.On CPU/GPGPU,this batch processing mode can improve the simulation speed by up to 5.02 or 34.29.Within this framework,comprehensive architectural exploration and end-to-end evaluation have been achieved,which provide some insights for systemic optimization.展开更多
文摘Efficient utilization of processor and memory resources is essential for sustaining performance and energy efficiency in modern computing infrastructures.While earlier research has emphasized CPU utilization forecasting,joint prediction of CPU and memory usage under real workload conditions remains underexplored.This study introduces a machine learning–based framework for real-time prediction of CPU and RAM utilization using the Google Cluster Trace 2019 v3 dataset.The framework combines Extreme Gradient Boosting(XGBoost)with a MultiOutputRegressor(MOR)to capture nonlinear interactions across multiple resource dimensions,supported by a leakage-safe imputation strategy that prevents bias frommissing values.Nested cross-validation was employed to ensure rigorous evaluation and reproducibility.Experiments demonstrated that memory usage can be predicted with higher accuracy and stability than processor usage.Residual error analysis revealed balanced error distributions and very low outlier rates,while regime-based evaluations confirmed robustness across both low and high utilization scenarios.Feature ablation consistently highlighted the central role of page cache memory,which significantly affected predictive performance for both CPU and RAM.Comparisons with baseline models such as linear regression and random forest further underscored the superiority of the proposed approach.To assess adaptability,an online prequential learning pipeline was deployed to simulate continuous operation.The system preserved offline accuracy while dynamically adapting to workload changes.It achieved stable performance with extremely low update latencies,confirming feasibility for deployment in environments where responsiveness and scalability are critical.Overall,the findings demonstrate that simultaneous modeling of CPU and RAM utilization enhances forecasting accuracy and provides actionable insights for cache management,workload scheduling,and dynamic resource allocation.By bridging offline evaluation with online adaptability,the proposed framework offers a practical solution for intelligent and sustainable cloud resource management.
基金supported in part by Beijing Academy of Artificial Intelligence(BAAI)(No.BAAI2019ZD0403)Beijing Innovation Center for Future Chip,Tsinghua Universitythe Science and Technology Innovation Special Zone Project,China。
文摘Resistive Random Access Memory(ReRAM)-based neural network accelerators have potential to surpass their digital counterparts in computational efficiency and performance.However,design of these accelerators faces a number of challenges including imperfections of the Re RAM device and a large amount of calculations required to accurately simulate the former.We present XB-SIM,a simulation framework for Re RAM-crossbar-based Convolutional Neural Network(CNN)accelerators.XB-SIM can be flexibly configured to simulate the accelerator’s structure and clock-driven behaviors at the architecture level.This framework also includes an Re RAM-aware Neural Network(NN)training algorithm and a CNN-oriented mapper to train an NN and map it onto the simulated design efficiently.Behavior of the simulator has been verified by the corresponding circuit simulation of a real chip.Furthermore,a batch processing mode of the massive calculations that are required to mimic the behavior of Re RAM-crossbar circuits is proposed to fully apply the computational concurrency of the mapping strategy.On CPU/GPGPU,this batch processing mode can improve the simulation speed by up to 5.02 or 34.29.Within this framework,comprehensive architectural exploration and end-to-end evaluation have been achieved,which provide some insights for systemic optimization.
