| Summary: | 碩士 === 國立中興大學 === 資訊科學學系 === 83 === Aiming at reducing the communication cost for distributed
query optimization, this thesis investigates two problems,
minimizing the communication cost for queries that refer to
n fragments (n>=2) and optimizing the execution orders of
large join queries. We define a pair - fragment allocation
matrix called PFAM and propose an algorithm to fill the
PFAM. The filled PFAM assures that any two fragments are both
allocated to at least one of the local sites, and all queries
that refer to two fragments can be executed just in the local
sites. The communication cost of all queries referring to
two fragments is thus reduced to zero and the total
communication cost in distributed query processing is also
reduced. A large join query consists of a series of join
operations. The order in which these joins are executed has
a great impact on the communication cost of distributed
query processing.We propose a dynamic programming algorithm to
optimize large join queries in distributed database
systems. By the algorithm, we can obtain the minimal
communication cost for some large join query as well as its
optimal parenthesization exec- ution orders.We can also
determine the sites to perform the query and the data fragments
to be transmitted among the sites.To demo- nstrate the research
results, the filled PFAM is applied to both the Distributed
INGRES optimization algorithm and the dynamic programming
algorithm for optimizing large join queries. It has been shown
that the communication cost is further reduced as compared
with that of the original two algorithms.
|
|---|
