Run Oracle OLTP workloads in HP-UX vPars and Integrity VM v6.1.5

Note:

It is not recommended to run any application at the VSP level, therefore the “native” mode tests are performed on an HP-UX server that

did not have the VSP configured.

As mentioned earlier in this paper, the Integrity BL890c i2 Server Blade has more physical memory than was intended for

the virtualized environment (256 GB vs. 64 GB); so, it was necessary to limit the blade’s visible memory during the native

mode test runs, to provide a balanced comparison. And, for this reason the memory on the Integrity BL890c i2 Server

Blade was restricted to 64 GB during boot for native mode runs and during vPars and Integrity VM guests' creation.

Results

Several different benchmark tests were run using Oracle database instance of 85 GB size, 38 GB of memory and 100

Swingbench users. Table 1 contains the results from the different Swingbench runs conducted during this effort. The

results in table 1 are an example of vPar/Integrity VM performance achieved during an internal benchmarking effort.

Your results will vary based on configuration and workload.

The columns in the table are as follows:

• Events: Uniquely identify each test configuration setup (with AVIO and DIO setup)

• Percentage of native performance with vPars: This indicates how close vPars is performing with respect to Native

• Percentage of native performance with Integrity VM Guest: This indicates how close Integrity VM Guest is performing

with respect to Native

Table 1.

Oracle/Swingbench benchmark results

Events Percentage of Native performance with

vPars Integrity VM Guest

TPS/AVIO 96.34% 82.84%

TPS/DIO 96.72% 83.64%

Percentages of native performance for HP-UX vPars and Integrity VM mentioned above are calculated from the TPS

numbers achieved in the virtualized environment and in native mode in the benchmark runs as reported in the respective

Oracle AWR reports. The computation is done as:

Percentage of native performance = (TPS in virtualized environment/TPS in native environment) * 100

Interpreting the results

Looking at the data in the above table, the following conclusions were made:

• AVIO performance is significantly improved in vPar providing near native performance. As the data indicates,

overhead with vPars new redesigned AVIO capability yields close to the same results as running on bare metal. Nearly

the same results were also observed with DIO for networking. The AVIO result reflects improvements made in the

redesigned AVIO storage subsystem for vPars v6.x.

Oracle workload deployed in a vPars v6.x configuration are best suited to high I/O usage as these types of workloads

also score the highest TPS numbers. While workloads deployed on Integrity VM are best suited to workloads that do not

have extreme I/O demands but need the continuous uptime benefits of VM High Availability feature such as online

migration and other features like dynamic memory and partial CPU entitlements.

DIO is relevant for networking only, showing noticeably better throughput when compared with AVIO in cases of

networking intensive benchmarks. Oracle benchmarks primarily stress disk IO operations and therefore, the numbers

shown in the AVIO and DIO results do not show a significant difference in observed overhead.