Query Processing for Edge Systems Accelerating Hash Joins using Workload Aware Column Imprints with Query Prediction

Abstract

quotThe number of edge devices in large-scale edge systems is rapidly increasing. newlineEdge devices have limited processing power, memory, and network bandwidth. newlineAs edge systems are resource-constrained, researchers have been trying to find newlineenergy-efficient query processing techniques that offer latency reduction. During newlinequery processing, joins have been one of the costliest database operations in terms newlineof both time and resources. newlineThe state-of-the-art solution for edge query processing, Column Imprint-Hash newlineJoin CI-HJ, is a data-aware approach that uses equi-height binning to accelerate newlinehash joins. It scans unnecessary cachelines, making it less energy-efficient. Because newlineof this, it cannot handle real-time query processing. The thesis presents newlineWorkload Aware Column Imprint-Hash Join WACI-HJ, which uses a workloadaware newlineapproach to accelerate hash joins. It computes the bins based on workload newlineinformation. Additionally, Workload Prediction is used to predict the upcoming newlinequery workload, which facilitates real-time query processing at the edge. newlineWACI-HJ comprises of (I) WACI-HJ Generation Phase (II) Query Processing newlineand Resource Utilization Phase. WACI-HJ Generation Phase has three modules: newline(1) Pre-processing (2) Prediction (3) Blocking and Hashing. In this phase, based newlineon the workload the bins are computed even before the query arrives, enhancing newlinethe algorithm performance. The Query Processing and Resource Utilization Phase newlineprocess queries and monitors CPU, RAM, and I/O utilization through (4) Query newlineProcessing (5) Resource Utilization modules respectively. newlineDuring the thesis work, challenges like Integrating Hash Joins with WACI, newlineIdentifying codes of Interest, Choosing an AppropriateWorkload Prediction Technique, newlineand Direct Measurement of Resource Utilization were addressed success newlinefully. newlineThe algorithm is demonstrated using benchmark and real-world datasets. For newlinemeasuring the algorithm performance, Percentage of Cache lines Read PCR, and newlineQuery Execution Time QET) is used.