Publication Type
Conference Paper
Abstract
Locking and timestamping are two popular approaches
to concurrency control in databases systems.
Although more than a dozen analytic performance
studies of locking techniques have recently appeared
in the literature, analytic performance study
of timestamp–based concurrency algorithms largely remains
an unexplored area. This work presents a model
of a distributed database system which provides the
framework to study the performance of timestamp ordering
concurrency control. We exhibit an analytical
solution, which has been tested with extensive simulation.
The accuracy seems to be very high. We assume
perfect and also imperfect clocks for synchronization
and quantify the way in which local clock inaccuracies
affect the phenomenon of transaction conflicts. In
particular, we derive a lot of interesting performance
measures such as probability of abort, throughput and
others.
to concurrency control in databases systems.
Although more than a dozen analytic performance
studies of locking techniques have recently appeared
in the literature, analytic performance study
of timestamp–based concurrency algorithms largely remains
an unexplored area. This work presents a model
of a distributed database system which provides the
framework to study the performance of timestamp ordering
concurrency control. We exhibit an analytical
solution, which has been tested with extensive simulation.
The accuracy seems to be very high. We assume
perfect and also imperfect clocks for synchronization
and quantify the way in which local clock inaccuracies
affect the phenomenon of transaction conflicts. In
particular, we derive a lot of interesting performance
measures such as probability of abort, throughput and
others.
Publication Links
Year of Publication
1993



