White Papers

27 Sizing and Best Practices for Online Transaction Processing Applications with Oracle 11g R2 using Dell PS Series | BP1003
Slightly higher IOPS (4,300 compared to 4,200) were observed with the ASM configuration. The IOPS
numbers maintained the generally accepted disk latency limit of 20 ms (for both read and write latencies
measured separately) for a random workload.
5.4 ORION/Vdbench on PS6110XS
The tests described in sections 5.3.1 to 5.3.3 were run with different capacity utilizations on the PS6110XS
array. Also, the ORION tool was used to simulate 8K-block-sized I/Os with 70/30 read/write mix which are
completely random in nature with no locality of data access.
The test results show that the array tries to use the SSDs first to provide the best performance. If the entire
database does not fit into SSDs, then the data is spread across both SSDs and 10K SAS drives. The
maximum IOPS on the storage array with 1 TB capacity utilization was 17, 000 IOPS. When the capacity
utilization on the array was increased to 2 TB, the maximum IOPS on the array was 10,230. One more
ORION test was run to simulate TPC-E like I/O transactions with heavy write intensive transactions and more
than 7,000 IOPS were observed as illustrated in section 5.3.3.
The decrease in IOPS as the capacity utilization increased is expected due to the large dataset that is
randomized and spread across more disk space. I/O generated by the ORION tool is completely randomized
and there is no mechanism to specify locality of data access, so there was not much performance
optimization to offer via the internal data tiering mechanism of the PS6110XS array.
However in typical OLTP database applications, most of the active data set is usually accessed frequently
and contained within a relatively small locality on disks. In these scenarios, the internal tiering feature of
PS6110XS array will identify the frequently accessed data and move the hot pages automatically to SSDs.
This can create significant performance improvements. The Vdbench I/O generation tool was used to
evaluate this scenario.
The Vdbench test was run to simulate 8K block sized I/Os with 70/30 read/write mix and 20% locality of I/O
access. When a specific locality of I/O got accessed frequently, it was clear that the PS6110XS array was
providing tiering benefits and the IOPS increased from 7,000 to 12,000 (refer to section 5.3.4 for more
details).
Additional tests were run as described in sections 5.3.5 and 5.3.6 to evaluate the benefits of Oracle ASM and
also to determine the volume configuration which produces optimal performance. Test results from sections
5.3.5 and 5.3.6 confirmed that using at least eight or more volumes per PS6110XS array member and ASM is
recommended for achieving optimal performance for Oracle OLTP database workloads.