Accelerating queries for structured and unstructured data

Jin, Tengjun

Permalink

https://hdl.handle.net/2142/124567 Copy

Description

Title

Accelerating queries for structured and unstructured data

Author(s)

Jin, Tengjun

Issue Date

2024-04-30

Director of Research (if dissertation) or Advisor (if thesis)

Kang, Daniel

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Database Systems
• Machine Learning
• Approximate Query Processing

Language

eng

Abstract

Data analytics is important for making data-driven decisions. As data volumes expand, the efficiency and cost of executing queries become critical concerns for analysts. Traditionally, analytics systems have prioritized structured data. Approximate Query Processing (AQP) systems, which provide faster aggregation queries by delivering approximate results, have been developed to enhance efficiency. However, they are limited used in real-application due to compatibility issues with popular databases and restrictions on the types of queries they can handle. To overcome these limitations, we have designed an innovative AQP system that functions as middleware. This system uses online sampling techniques to accelerate aggregation queries and can meet user-specified error targets. With advancements in machine learning (ML), analysts are increasingly interested in analyzing unstructured data (videos, images, text, and audio) to extract semantic information. Current analytics systems typically integrate ML models through user-defined functions (UDFs). These UDFs can be difficult to optimize and require application users to write complex, nested table expressions. To address these challenges, we introduce a new data model, AIDM, enabling users to query ML model outputs as standard SQL tables, through virtual columns and virtual tables. We implement AIDM, as well as novel optimizations for accelerating both approximate and exact queries in AIDB. Our evaluations show that the AQP system can provide speedups of up to 87x and AIDB can reduce the number of ML model invocations by up to 98%.

Graduation Semester

2024-05

Type of Resource

Text

Handle URL

https://hdl.handle.net/2142/124567

Copyright and License Information

Copyright 2024 Tengjun Jin

Owning Collections