Next: Uniform Workloads
Up: Performance Model
Previous: Stage 2: Joining
Before designing techniques for partitioning temporal data we want to
evaluate some characteristics of the performance model that was
described in sections 8.2 - 8.4.
Interesting questions, for example, are:
- Which cost components dominate? Is it I/O, communication,
CPU time or memory accesses?
- What is the ratio between and ?
- Which workload parameters affect which cost components?
The problem about answering these questions is that the
partition that underlies the simulated join computation is an
important factor within the cost model. Therefore it is impossible to
evaluate the entire performance model without making any assumptions
about the underlying partition. In order to reduce the importance of
that we will create experiments for computing temporal intersection
joins
assuming that the underlying data is uniform.
Section 8.5.1 describes the uniform workload. In
section 8.5.2, this workload is used to
perform a series of experiments that provide useful information about
the characteristics of the performance model.
Thomas Zurek