智能边缘计算
文章背景与核心概要
在大规模系统中边缘设备的快速激增，由于这些设备有限的处理能力、内存和网络带宽，给资源利用和数据管理带来了巨大挑战。数据库连接（Database joins）作为代价最高昂的数据库操作之一，进一步加剧了这些局限性。虽然当前最先进的边缘查询处理方法（如列印痕-哈希连接 CI-HJ）使用等高分箱（equi-height binning）来加速连接，但它们在实时处理方面效率低下，并且经常扫描不必要的缓存行（cachelines）。
为了克服这些障碍，本文介绍了工作负载感知列印痕-哈希连接（Workload Aware Column Imprint-Hash Join, WACI-HJ）。通过提前预测即将到来的查询工作负载，WACI-HJ 优化了实时边缘查询处理。它通过两个主要阶段运行：1. WACI-HJ 生成阶段：结合了预处理、预测和分块/哈希模块，在查询到达之前根据预测的工作负载计算分箱。2. 查询处理与资源利用：管理活动查询处理以及 CPU、RAM 和 I/O 利用率。
在基准数据集和真实世界智能交通数据集上的评估表明，WACI-HJ 减少了 54% 的读取缓存行，并将查询执行时间提高了 10%。此外，它在各项资源上都实现了效率提升（CPU 提升 1%, RAM 提升 38%, I/O 提升 49%），并直接优化了实时交通分析、拥堵管理和路线规划应用。
摘要
The rapid proliferation of edge devices in large-scale systems introduces significant challenges regarding resource utilization and data management due to these devices' limited processing power, memory, and network bandwidth. Database joins—among the most expensive database operations—further exacerbate these limitations. While current state-of-the-art edge query processing methods like Column Imprint-Hash Join (CI-HJ) use equi-height binning to accelerate joins, they suffer from inefficiencies in real-time processing and frequently scan unnecessary cachelines.
To overcome these hurdles, this paper introduces the Workload Aware Column Imprint-Hash Join (WACI-HJ). By predicting upcoming query workloads in advance, WACI-HJ optimizes real-time edge query processing. It operates through two primary phases: 1. WACI-HJ Generation Phase: Incorporates pre-processing, prediction, and blocking/hashing modules to compute bins based on predicted workloads before queries arrive. 2. Query Processing and Resource Utilization: Manages active query processing alongside CPU, RAM, and I/O utilization.
Evaluated on benchmark datasets and real-world Smart Transportation datasets, WACI-HJ demonstrates a 54% reduction in cachelines read and a 10% improvement in query execution time. Furthermore, it achieves efficiency gains across resources (1% in CPU, 38% in RAM, and 49% in I/O) and directly optimizes real-time traffic analysis, congestion management, and routing applications.
元数据与参考信息
- arXiv ID: arXiv:2609.00181 [cs.DB]
- DOI: 10.48550/arXiv.2609.00181
- Primary Subject: Databases (
cs.DB)- Secondary Subject: Artificial Intelligence (
cs.AI)- Authors:
- Kalgi Gandhi
- Minal Bhise
- Submission Date: 31 August 2026
链接与全文访问
