Databases Reference
In-Depth Information
from other applications on both machines, only the basic processes needed by the
operating system, network, I/O, and the processes needed for the benchmark are
running.
The databases analyzed in the following section, are commercial products
used in industry for daily business in different areas. Three DBMS with differ-
ent data storage characteristics (as discussed in Sect.
2.2.4
) are tested: column-
oriented in-memory (CM), row-oriented disk-based (RD), and column-oriented
disk-based (CD).
Results shown in this chapter are normalized to
per DBMS with 1 marking
the longest response time in the result set or the highest throughput. The reason
for this normalization is that the focus of the evaluation lies in the application of
the proposed benchmark to make observations on behavioral aspects of database
types under the variation of workload and database schemas. It is not in the scope
of this evaluation to compare the actual performance of different database systems
with each other based on the results. Basic optimizations are applied. These include
standard indexes on primary key attributes, on foreign key attributes, and on further
attributes used in
joins
,
where
,and
group by
conditions.
Œ0 W 1
7.2
Impact of Adding OLAP to OLTP
The goal of this analysis is to evaluate the impact of combining the transactional
with the analytical workload. The following questions are posed and answered in
this section:
1. What is the impact on query performance when adding analytical queries to the
workload, that is, how does the performance of transactions change and to what
degree does it decrease?
2. Do databases show different behavior depending on their storage types (row/
column-orientation, in-memory/disk-based)?
Simulated workload mixes contain the following combinations of the number of
OLTP clients and OLAP clients:
f1; 5; 10; 50; 100gf1; 5; 10; 50; 100g
Figure
7.1
provides an excerpt of the results of this analysis for the tested DBMS.
These tests use the original database schema variant of CBTR in 1NF. The charts on
the left depict how throughput behaves when adding OLTP and OLAP clients. The
ones on the right show how the response time of mixed OLTP queries changes. The
series in both graphs mark the configured base workload of chosen configurations
that include between 1 and 100 OLTP clients (legend given below the graphs)
running in parallel. The number of OLAP clients specified on the x-axis is added to
these. See Table
C.1
for a tabular overview of the results.
The measured throughput in all systems increases with the addition of OLTP
clients because of sending a larger number of short requests to the database.
Search WWH ::
Custom Search