DBMbench:
Microbenchmarking Database Systems in a Small, yet Real World
Tuesday September 16, 2003
3rd floor Intel Research Pittsburgh (417 S. Craig Street)
4:00 pm
Minglong Shao
Carnegie Mellon University
Database benchmarks are widely used to test functionality and evaluate
the overall performance of database systems. However, the complexity
involved in establishing the experimental environment and in conducting
detailed analysis makes conventional benchmarks inappropriate for use
in research that evaluates low-level hardware, such as processor microarchitecture
and memory hierarchy performance behavior of database systems. It is
often necessary to scale down benchmark parameters such as dataset size
or query complexity. Scaling down is only valid, however, if the new
benchmarks preserve the typical characteristics of the original benchmark
for the purpose that they are used. In this talk, I will introduce a
systematic scaling framework to scale down conventional benchmarks and
present extrapolation rules to predict the processor and memory behavior
of full-scale benchmarks. I will also propose the DBMbench consisting
of two substantially scaled-down microbenchmarks: MTPC-H and MTPC-C that
capture the processor and memory performance behavior of DSS and OLTP
workloads.
Minglong Shao is a Ph.D. candidate in the Department of Computer Science
at Carnegie Mellon University. She received her B.A. from the Department
of Computer Science and Technology, Tsinghua University. Currently, she
is working with Professor Anastassia Ailamaki on database performance
characterization. Her research focuses on techniques to improve database
performance throughout the memory hierarchy.
|