Scholarlink

<p>Continuous range queries (CRQs) involve persistent queries on streaming data that continuously retrieve new records that fall within specified ranges. Due to high memory overheads and time costs, traditional query indexing structures have difficulty handling large-scale continuous-range query data. To address this problem, we propose a learned index for CRQs over streaming data (LICRQ). The LICRQ combines a multilayer grid, spatial range filters, and learned indexes to accelerate the processing of numerous CRQs of various sizes while reducing memory consumption. We also introduce the G-LICRQ, a GPU-accelerated extension of the LICRQ that significantly improves the processing throughput. The experimental results show that the LICRQ outperforms CKDB-tree, achieving approximately 50% faster query times and 20% lower memory usage, and surpasses PriPL-Tree with approximately 23% faster queries and 16% less space overhead than traditional indexes. Furthermore, the LICRQ decreases the query overheads by 12% compared with the learned index FLIRT, and the G-LICRQ reduces the query time by approximately 66% compared with that of the LICRQ and by approximately 43% compared with that of G-CKDB-tree.</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

A learned index for continuous range queries over streaming data

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Ze Deng,
  • Xiang Li,
  • Lizhe Wang,
  • Wei Qi

摘要

Continuous range queries (CRQs) involve persistent queries on streaming data that continuously retrieve new records that fall within specified ranges. Due to high memory overheads and time costs, traditional query indexing structures have difficulty handling large-scale continuous-range query data. To address this problem, we propose a learned index for CRQs over streaming data (LICRQ). The LICRQ combines a multilayer grid, spatial range filters, and learned indexes to accelerate the processing of numerous CRQs of various sizes while reducing memory consumption. We also introduce the G-LICRQ, a GPU-accelerated extension of the LICRQ that significantly improves the processing throughput. The experimental results show that the LICRQ outperforms CKDB-tree, achieving approximately 50% faster query times and 20% lower memory usage, and surpasses PriPL-Tree with approximately 23% faster queries and 16% less space overhead than traditional indexes. Furthermore, the LICRQ decreases the query overheads by 12% compared with the learned index FLIRT, and the G-LICRQ reduces the query time by approximately 66% compared with that of the LICRQ and by approximately 43% compared with that of G-CKDB-tree.