|
|
Thesis: "Goal Oriented
Buffer Pool Tuning Algorithms for Decision Support Workloads"
|
|
Abstract (full text
not available)
|
|
Database Administrators (DBAs)
manage the performance of the database management systems (DBMSs) by manually
adjusting low-level system parameters. The process of configuring
and tuning DBMSs is both complex and time-consuming. As the complexity
and diversity of data types and database workloads increase, manually tuning
by DBAs is becoming almost impossible.
More and more database experts agree that moving the responsibility for
tuning the DBMSs away from the DBA onto the system itself will be necessary
in the next generation of DBMSs. The DBMS, therefore, would automatically
tune low-level configuration parameters based on pre-defined high-level
performance goals provided by the DBA.
A self-tuning DBMS needs algorithms to manage its resources. These
algorithms detect the violations of system performance and dynamically reallocate
resources to improve the performance of the system. This thesis presents
a self-tuning algorithm for a key resource in a DBMS, namely its buffer
pool. The algorithm is aimed at tuning large complex queries that
involve multiple joins, which are typical of decision-support workloads.
It uses the information from the query optimizer to guide buffer allocation.
The performance of the algorithm is evaluated by a set of experiments
with DB2 Universal Database [IBM] and compared with another self-tuning
algorithm, the Dynamic Reconfiguration Algorithm [LI99].
|
|