Document Type

Campus-Only Access for Five (5) Years

Embargo Period

2-1-2017

Degree Program

Electrical & Computer Engineering

Degree Type

Master of Science (M.S.)

Year Degree Awarded

2016

Month Degree Awarded

September

Advisor Name

Lixin

Advisor Last Name

Gao

Abstract

In the era of big data, people not only enjoy what massive information brings, but also experience the problem of information overload. As the volume of both data and users increasing sharply, more and more studies focus on how to answer a query for interesting information from massive data. However, most memory-based query systems are designed and implemented to optimize the performance in processing a single query and do not support in-memory data sharing among query processing jobs. When they are extended to process multiple concurrent queries, they will suffer the problems of the inefficient use of memory and waste of time.

This thesis aims to design and implement a memory-efficient system, ParQ, which can be adopted by memory-based query systems to realize query-level parallelism. The main idea includes constructing a common memory block for maintaining sharable data. By sharing data, ParQ is able to process multiple queries concurrently while reducing memory usage and running time. We apply ParQ to several existing query systems. The experiment results show that ParQ improves the performance in both job completion time and memory usage when executing multiple concurrent query jobs.

Share

COinS