Scholarlink

<p>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 conduct a comprehensive analysis of the attributes associated with various GPU requests and allocations in heterogeneous clusters. We introduce the adaptive selection model (ASM), which dynamically selects the best-performing classifier to predict the demand for different types of GPUs. Additionally, we utilize Gini importance to identify the most significant features that bridge the gap between requests and allocations. Furthermore, we implement spatial and temporal long short-term memory (ST-LSTM) to enhance the prediction of GPU type allocation. We validate the effectiveness of our proposed methodology using real-world trace data from Alibaba Cloud data centers. The results indicate that the ASM achieves an impressive accuracy rate of 87%, while the ST-LSTM model demonstrates excellent performance with average RMSE and MAE values of 1.84 and 1.09, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting GPU type allocation via demand feature extraction in heterogeneous clusters

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Sheng Wang,
  • Yumei Shi,
  • Shiping Chen,
  • Meng Liu

摘要

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 conduct a comprehensive analysis of the attributes associated with various GPU requests and allocations in heterogeneous clusters. We introduce the adaptive selection model (ASM), which dynamically selects the best-performing classifier to predict the demand for different types of GPUs. Additionally, we utilize Gini importance to identify the most significant features that bridge the gap between requests and allocations. Furthermore, we implement spatial and temporal long short-term memory (ST-LSTM) to enhance the prediction of GPU type allocation. We validate the effectiveness of our proposed methodology using real-world trace data from Alibaba Cloud data centers. The results indicate that the ASM achieves an impressive accuracy rate of 87%, while the ST-LSTM model demonstrates excellent performance with average RMSE and MAE values of 1.84 and 1.09, respectively.