Introduction - Benchmarking Transaction and Analytical Processing Systems - page 8

• What are typical logical database optimizations? The optimized database

schemas of transactional and analytical processing represent the opposite ends of
a scale. While transaction processing schemas are optimized for high throughput
of write operations and highly selective read operations, analytical systems are
optimized for company-wide information retrieval requirements reading large
amounts of data. In a hybrid setting, the database schema has to be optimized
according to the workload mix and prioritized operations.

• What is the influence of database schemas that utilize different optimizations?

This influence on queries has to be determined to support making a decision for
specific schema optimizations. A certain schema optimization can have a positive
impact on the execution of a subset of queries of the entire mixed workload while
impeding the performance of other queries. The quantification of this impact and
its understanding is the first step towards making a decision on using a particular
optimization in a mixed workload scenario.

• What are additional factors that influence the performance of database design

in a mixed OLTP and OLAP environment? Workload mix is not the only factor
influencing the decision of which database schema is best suited in a specific
setting. The type of database, that is, its storage layout, is another important factor
of influence. In the OLAP environment, both column and row-oriented physical
storage layouts are common, while row-oriented storage is still the prevailing lay-
out of databases underneath OLTP applications. The same schema optimizations
on top of different storage layouts may have different impact on queries.

1.2 Contribution and Scope

According to the questions identified in the previous section, the contributions of
this thesis are in the areas of benchmarking and understanding logical database
design decisions in mixed OLTP and OLAP workload scenarios. The contributions
and adjacent works have been published in international conferences, workshops,
and journals.

Benchmarking

Existing benchmarks targeting relational databases in OLTP or OLAP environments
are compared and discussed with regard to observations from current enterprise
workloads. Enterprise workloads have evolved since the creation of the benchmarks
widely in use today. Yet, the benchmark workloads remained the same since their
specification with minor changes to requirements.
The review of existing benchmarks and observations of current enterprise
systems provide the foundation for the definition of the composite benchmark for
transactions and reporting (CBTR) proposed in this thesis. CBTR is based on a real
database design and it provides realistic queries to simulate different workloads.