Principles of Instruction-Level Distributed Processing

The clustered machines, by contrast, are shown to be inherently capable of matching monolithic machine performance, the penalties imposed by distributed execution notwithstanding. Key to exploiting that potential is knowledge of the critical path through a program. This can be used to achieve a judicious allocation of execution resources to instructions, with performance-critical instructions being shielded from the distributed machine's execution constraints; only the least important instructions, which can tolerate some delay, need be exposed to those constraints. This dissertation develops several novel critical path-aware schemes, and shows that they can deliver performance that is within a few percent of a monolithic machine. It further shows that many aspects of those schemes are stable, both within and across runs of a program, a property which lends them to implementation in a static (offline) context.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Salverda, Pierre M.

Contributors dc:contributor

• Zilles, Craig

• Computer Science

Language dc:language

eng

Identifier

(MiAaPQ)AAI3314880

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/81815