| Summary: | 碩士 === 國立臺灣大學 === 資訊工程研究所 === 81 === A global time base has been widely recognized as an important
requirement in distributed computing systems. It can be used to
simplify the solutions to several design problems such as
check- -pointing, interprocess communication, resource
allocation, and transaction processing, especially in the
presence of faults. A global time base can be established by
synchronizing all the local clocks in the distributed system.
This synchronization would not have been much of a problem had
all the clocks includ- -ing the faulty ones behaved
consistently with one another. How- -ever, when some of the
faulty clocks behave in any arbitrary manner, clock
synchronization can pose some serious problems. For instance,
a faulty clock can make it difficult for the other clocks to
synchronize themselves by either omitting or sending
conflicting clock information during the course of
synchroniza- -tion. The kind of arbitrary behavior is referred
to as a Byzan- -tine fault. In this thesis we present an
integrated and simple scheme to the problems of synchronizing
clocks, initializing these clocks, and integrating new or
repaired clocks, for distributed systems with different types
of failures that are likely to occur during a system's
lifetime, such as omission faults and performance faults
that do not partition the communication network. The scheme
proposed here is particularly suitable, but not limited, for
large partially-connected distributed systems.
|
|---|
