Skip navigation
Please use this identifier to cite or link to this item:
Title: Runtime Speculative Software-Only Fault Tolerance
Authors: Zhang, Yun
Advisors: August, David I
Contributors: Computer Science Department
Keywords: Fault Tolerance
Subjects: Computer science
Issue Date: 2012
Publisher: Princeton, NJ : Princeton University
Abstract: Transient faults are emerging as a critical reliability concern for modern microprocessors. Recently, microprocessors have been designed with lower voltage level ,smaller and faster transistors enabled by improved fabrication technology. A combination of increased density of transistors on chip, reduced noise margin of each transistor, and voltage scaling are making hardware systems more susceptible to transient faults than ever. Both hardware or software solutions have been proposed for transient fault tolerance. The hardware approach typically adds redundant hardware modules to the system, thus requiring extra chip area as well as higher hardware design and verification cost. In ad- dition, the scope and mechanism of fault tolerance are hardwired at design time, which could be suboptimal with the change of deployment environment. Unlike hardware solu- tions, software-only techniques do not require any specialized hardware extensions and are more flexible with the scope of protection and the change of environment. However, even the best-performing software-only fault tolerance techniques incur significant performance cost. The overhead of prior work comes from doubled register usage, frequent inter-core communication, or barrier synchronizations. These factors prevent existing software tech- niques from being adopted widely. To address these problems, this dissertation proposes Runtime Software-only Speculative Fault Tolerance (RSFT). The key insights behind this dissertation are: (1) not all values are equally important. Transient faults may alter a transistor's value, which is never used. Only the values that will affect the externally visible behavior of a program must be verified before being used; (2) Value speculation can efficiently remove data dependences introduced by cross checking values produced in the program and its redundant copy with high confidence, thus significantly improves program runtime performance. RSFT serves as a virtual layer between the application and the underlying platform.
Alternate format: The Mudd Manuscript Library retains one bound copy of each dissertation. Search for these copies in the library's main catalog
Type of Material: Academic dissertations (Ph.D.)
Language: en
Appears in Collections:Computer Science

Files in This Item:
File Description SizeFormat 
Zhang_princeton_0181D_10157.pdf803.27 kBAdobe PDFView/Download

Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.