Title

Near Data Processing for Clustering Enhanced by Automatic Memory Disaggregation

DGIST Authors

Sanghoon Lee ; Yeseong Kim ; Sungjin Lee

Advisor

김예성

Co-Advisor(s)

Sungjin Lee

Issued Date

2023

Awarded Date

2023-02-01

Citation

Sanghoon Lee. (2023). Near Data Processing for Clustering Enhanced by Automatic Memory Disaggregation. doi: 10.22677/THESIS.200000656260

Type

Thesis

Description

Clustering Algorithms, Near Data Processing, Disaggregated Memory, CXL, Genetic Algorithms

Table Of Contents

Ⅰ. INTRODUCTION 1
II. RELATED WORK 6
III. OVERVIEW OF SIDEKICK 8
3.1 Architectural Overview: Memory Disaggregation with CXL 10
3.2 Core Software: Automated Migration Technique 12
IV. AUTOMATED PROGRAM CONTEXT EXTRACTION 13
4.1 Program Context Extraction 13
4.2 Managing computation and memory allocation 15
4.3 Implementation 16
V. MIGRATION POLICY DECISION 17
5.1 Initialization of Genetic Algorithm 18
5.2 Identifying the best migration policy using genetic algorithm 19
VI. EXPERIMENTAL RESULTS 21
6.1 Experimental Setup 21
6.2 Migration Decision Evaluation 26
6.2.1 Effectiveness and Scalability 26
6.2.2 Migration Quality Evaluation 27
6.3 Memory Activity Evaluation 29
6.4 In-Depth Analysis of Sidekick Optimization Procedure 30
6.5 Impact of Bootstrapping 31
VII. CONCLUSION 33

URI

http://hdl.handle.net/20.500.11750/45760
http://dgist.dcollection.net/common/orgView/200000656260

DOI

10.22677/THESIS.200000656260

Degree

Master

Department

Department of Electrical Engineering and Computer Science

Publisher

DGIST

Show Full Item Record