|
|
Thesis: "A Self-Tuning
Algorithm for Multiple Buffer Pools in Database Management Systems"
|
|
Abstract (full text
not available)
|
|
Tuning database management
systems (DBMSs) requires knowledge of the characteristics of the systems,
the data, and the workload, and of the interrelationship among them. Due
to the increasing complexity of the DBMSs and diversity of their workloads,
there is a need for automatic tuning of DBMSs.
The buffer pools are a key resource in a DBMS; their existence reduces the
number of disk accesses performed by a transaction. The size of each
buffer pool is set by configuration parameters and page replacement is local
to each buffer pool. Tuning the size of buffer pools is crucial to achieving
good performance.
Dynamic Reconfiguration algorithm (DRF) [Li99] is a self-tuning algorithm
for multiple buffer pools in a DBMS. It implements the concept of goal-oriented
resource management by reallocating the sizes of buffer pools to meet the
performance goals of various transaction classes. It is an iterative
algorithm that uses greedy heuristics to find a suitable buffer allocation.
However, it adopts a relatively simple database system model, which
only takes into account the impact of a few factors.
This thesis presents an improved DRF by developing a new access cost equation
and a new miss rate estimator. We implemented the improved DRF on DB2
Universal Database. The performance of the algorithm is experimentally
studied for both TPC-C and DBMIN workloads and compared with the original
DRF.
|
|