Tr-Predictior: An Ensemble Transfer Learning Model for Small-Sample Cloud Workload Prediction.

Area of Science:

• Cloud Computing
• Machine Learning
• Data Science

Background:

• Accurate workload prediction is crucial for intelligent scheduling in cloud platforms.
• Short-workload sequences present challenges due to small data volumes and outliers.
• Existing methods struggle with the unique characteristics of short-workload data.

Purpose of the Study:

• To propose an effective method for predicting short-workload sequences in cloud environments.
• To address the challenges of small data size and outliers in workload prediction.
• To enhance the accuracy of workload prediction for intelligent cloud scheduling.

Main Methods:

• Developed Tr-Predictor, an ensemble learning method combining sample weight transfer and Long Short-Term Memory (LSTM).
• Utilized Time Warp Edit Distance (TWED) and Transfer Entropy (TE) for similar sequence selection.
• Employed a two-stage weight adjustment strategy based on sample and model errors, upgrading the learner to LSTM.

Main Results:

• Tr-Predictor demonstrated higher prediction accuracy compared to existing methods on small-sample workloads.
• Experimental validation using Google and Alibaba cluster datasets confirmed the algorithm's effectiveness.
• Ablation experiments highlighted the performance gains contributed by individual components of the proposed method.

Conclusions:

• The proposed Tr-Predictor method effectively improves short-workload sequence prediction accuracy in cloud platforms.
• The combination of sample weight transfer, LSTM, and optimized weight adjustment is key to the method's success.
• This approach offers a robust solution for intelligent scheduling in data-scarce cloud environments.