Data Storage vendors always (or should) want to ensure that they build systems that solve customers’ application requirements and issues. As a leading provider of flash memory data storage systems, Tegile Systems is no different. Additionally, since our storage platform is being “built for speed,” we continually conduct performance benchmarks to ensure that we have a competitive advantage over “the other guys”.
Above is a Magic Quadrant (MQ) from Gartner, a leading Technology Research firm, which shows the relative market positioning of Database Management Systems. So, any storage provider selling to enterprise customers needs to take Oracle Database requirements seriously. Inevitably, performance moves to the forefront of these discussions.
Deployment of flash arrays eliminates many of the traditional chokeholds seen in Oracle Database environments – storage is no longer at the root for every Oracle performance bottleneck. However, this creates new challenges in quantifying and resolving Oracle Database performance from a storage context.
Recently, we conducted a test to measure the performance of a typical application workload running on an Oracle Database environment.
- To measure performance, we used the graphical open-source HammerDB benchmarking tool. This tool is readily available, widely used, and relatively easy to deploy on Windows or Linux. HammerDB became an obvious choice because it included a workload derived from the industry-standard TPC-C benchmark.
- We chose the HammerDB’s TPC-C workload generator to simulate an On-Line Transaction Processing (OLTP) I/O pattern. A TPC-C workload mimics a wholesale supplier with sales districts and associated warehouses. The operational business flow includes order entry, order processing, order delivery, payment recording, order status checking, and warehouse inventory management.
- NOTE: Official Audited TPC-C benchmarks are extremely costly, time consuming and complex to establish and maintain. HammerDB takes the essence of this benchmark and implements it in a way that can be run at a fraction of the cost. Faster and cheaper holds great appeal.
- Oracle Database 12c Enterprise Edition with Oracle Grid running on Oracle Linux
- Dell PowerEdge Server and Brocade Switch
- Tegile IntelliFlash T4700 All-Flash Storage Array
We achieved an impressive result of more than four million transactions per minute (TPM)!
A Few Notes About Measuring Oracle Database Performance
Although these results are astounding, Tegile IntelliFlash storage arrays are capable of much greater feats!
For all its pluses, the big limitation of HammerDB is that it doesn’t adequately stress modern storage subsystems. Inevitably, bottlenecks in CPU and concurrency are realized before storage ever becomes a factor. In our testing, the database server CPU became saturated with less than 5000 IOPS – Tegile all-flash arrays are capable of delivering more than one million IOPS under the right conditions.
As storage systems move from hard drives to flash to persistent memory, benchmarking strategies much also shift. Rather than ensuring that storage does not slow down Oracle database performance, we must tune differently so that Oracle Database environment does not slow down storage performance.
Typically, Tegile all-flash arrays can do reads from flash memory in microseconds. However, Oracle databases must do additional work to ensure read consistency and ensure data integrity – all of that takes additional server CPU cycles. Therefore, the challenge in these benchmark exercises is to get Oracle out of the way of itself so that storage performance can really shine. That is a non-trivial task. The CPU bottleneck is confirmed in Oracle AWR reports which list CPU as the top wait events.
Large data block buffers shift the focus from physical I/O operations to logical I/O operations. Sub-optimal SQL tuning may cause excessive logical I/O. Library cache should also be investigated to determine if excessive parsing might be increasing CPU utilization. Assuming an optimized Oracle database environment, the remaining options are to add more CPUs or faster CPUs.
While four million transactions per minute is impressive, the true capabilities of Tegile storage will be highlighted by adding additional database servers and Oracle instances to our HammerDB configuration. Tegile’s active/active controller capabilities make this expansion an easy task.
Download the Results
Check out our latest results for Oracle Database 12c Performance on the Tegile IntelliFlash T4700 all-flash storage array: