Architecture comparison

Runtimes are compared for the IBM® System z9® and IBM System z10™ architectures.

We imported an export file of 20 GB containing data from multiple table spaces into an Oracle database with explicit index creation. We used a z/VM® guest with 2 CPUs and 16 GiB memory (see Table 1) on a System z9 and on a System z10 and measured the required time.

The import was initiated using the following statement
imp dbuser/passwd PARFILE=PARFILE.imp FILE=/loaddata/loadtest.dmp
with the following import configuration file (PARFILE.imp):
BUFFER=5242880
SHOW=N
IGNORE=Y
GRANTS=Y
INDEXES=Y
COMMIT=Y
CONSTRAINTS=Y
FROMUSER=(dbuser)
TOUSER=(dbuser)
The resulting runtimes are shown in Figure 1, normalized that the z9® runtimes are 100%.
Figure 1. Execution times required for importing 20 GB data and index creation on System z9 and System z10

Execution times required for importing 20 GB data and index creation on System z9 and System z10

The chart shows that this process needs only half the time on System z10 than on System z9, which is an excellent improvement for this scenario.

During the import phase, we observed a disk I/O rate of about 70 MB/sec (read and write), which can be considered as a high load for database I/O. During index creation, the disk I/O rate was about 250 MB/sec (read), which is a really heavy load. Both values show the strengths of the I/O setup with the separation of the I/O patterns and the striped storage pools on the DS8000®.