Transparent process rollback recovery : some new techniques and a portable implementation /

deterministic execution, but is disabled and replaced with

Bibliographic Details
Main Author: Ellenberger, Ernest Lloyd, 1970-
Format: Thesis eBook
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1995.
Subjects:
Online Access:Link to OAKTrust copy
Description
Summary:deterministic execution, but is disabled and replaced with
deterministic processes by using causal dependency
execution to be deterministic. Research prototypes have
family-based message logging with the logging site
family-based message logging. If a failure occurs, the
Finally, the concept of reactive replication for message
immediately copied to stable storage or to the volatile
implementation has not been readily available. To address
implementation, and failure-free performance of a new
information that makes a distinction between
logged data necessary to recover the failed process is
logging implementations to be reasonably small for a
logging is introduced. During failure free operation,
Marzullo [4]). Performance measurements show the overhead
nondeterministic and deterministic state intervals.
nondeterrninistic. The approach requires the application
of the consistent checkpointing and family-based message
optimistic checkpointing during periods of nondeterminism.
overhead for every message sent, and requires process
overhead, or message logging, which introduces a smaller
period of nondeterminism. Message logging is used during
periodically introduces a relatively large but temporary
presents a new approach for efficient output commit and
Processes in a distributed system can be made transparently
processes share a common memory address space (recently
program to explicitly mark the beginning and end of each
reactive replication uses a low-overhead protocol that can
recoverable through the use of process checkpointing, which
recovery when some processes are intermittently
representative distributed application. This thesis also
shown promising performance results [1, 21, but an
storage of another processor; a second failure can be
suggested independently by Vaidya [3) and Alvisi and
technique for reducing message logging overhead when some
than the recovery protocol.
that deficiency, this thesis describes the design,
The commit algorithm avoids communication with
The system incorporates both coordinated checkpointing and
This technique assumes that such a copy operation is faster
tolerate only a single simultaneous failure, such as
tolerated only after the completion of that copy operation.
transparent recovery system for standard Unix workstations,
which provides a basis for future experimental work.
Item Description:"Major subject: Computer Science".
Vita.
Physical Description:xi, 99 leaves : illustrations ; 28 cm.
Also available online.
Issued also on microfiche from Lange Micrographics.
Bibliography:Includes bibliographical references.