Acceleration of Relational Database Operations on Vector
Processors
Shintaro Meki
Faculty of Computer Science & Systems Engineering, Okayama Prefectural University, Soja, Japan 719-1197
Yahiko Kambayashi
Graduate Faculty of Engineering, Kyoto University, Kyoto, Japan 606-8501
SUMMARY
With the expansion of the areas of database applica-
tions, the amount of data processed has become gigantic,
and the demand for faster database systems has increasingly
intensified. Thus, many efforts have been expended to
execute database operations more efficiently using special
hardware. However, for the cases in which mass-produced
LSI cannot be used, cost has become excessive, and replace-
ment methods have been sought. Vector processors have
powerful pipeline operators and large-capacity main
memories and have rapidly spread in recent years. In this
paper, algorithms for relational database operations suitable
for vector processors showing the characteristics of vector
processors sufficiently are presented. This algorithm is
based on the hash method, which is highly efficient when
the capacity of the main memory is large, and it is shown
that a rate of acceleration more than 10 times the average is
obtained for representative relational database operations
when this algorithm is executed on a vector processor. © 2000
Scripta Technica, Syst Comp Jpn, 31(8): 7988, 2000
Key words: Vector processor; main memory data-
base; join algorithm; hash technique.
1. Introduction
The demand for high-speed database management
systems (DBMS) has increased each year due to the expan-
sion of areas of database applications in recent years. Thus,
special hardware for database operations has generally been
developed. However, since such trial systems have been
expensive and much time is required to design such LSI,
the hardware technology would become antiquated when
the hardware is completed. Two methods are comparatively
successful approaches regarding the development of
DBMS satisfying the cost requirements: (1) use of a com-
puter with a large-capacity main memory for executing a
main memory database and (2) use of many microproces-
sors for executing parallel database systems.
The cost of memory devices has decreased each year
with the advancement of the technology for large-scale
integrated circuits, and the capacity of the main memory for
a computer has become increasingly large. Thus, all major
parts of data can now be placed in the main memory, and
thus, high-speed DBMS with a decreased number of ac-
cesses to secondary memory devices have become possible.
In addition, with the advancement of hardware technology,
the capabilities of microprocessors have increased while the
cost has decreased, making economical realization of par-
allel database machines possible. Since these approaches
use mass-produced LSI chips, special LSI for database
© 2000 Scripta Technica
Systems and Computers in Japan, Vol. 31, No. 8, 2000
Translated from Denshi Joho Tsushin Gakkai Ronbunshi, Vol. J81-D-I, No. 8, August 1998, pp. 976985
79