Record and replay under relaxed consistency

In the area of debugging parallel executions, record and replay is a technique that allows deterministic debugging even in the presence of data races. It is useful as most programmers are used to re-executing programs to find bugs. However, very little is known about how the consistency model affects record and replay. Previous work only applied to very strong consistency models, or to a specific architecture of shared memory. Very little theoretical basis has been developed for record and replay. This thesis makes three contributions: • An algorithm that records the minimum record for record and replay under causal consistency. • A demonstration that guaranteeing progress for a given replay mechanism can depend on the consistency model. • A demonstration that heterogeneous consistency record and replay is possible, that is, it is possible to record an execution on one consistency model and replay it on another.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2016

Author dc:creator

• Jones, Russell Llewellyn

Contributors dc:contributor

• Vaidya, Nitin

• Record and Replay
• Causal Consistency
• Memory Consistency
• Distributed Memory
• Distributed Computing

Statement dc:rights

• Copyright 2016 Russell Jones

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/92729

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/92729