Home > Workload Solutions > SQL Server > White Papers > Dell EMC PowerMax Storage for Mission-Critical SQL Server Databases > Test case 5: Noisy neighbor test
This test demonstrates how PowerMaxOS maintains the performance levels needed for a mission-critical application by managing a “noisy neighbor” – the temporary but demanding application competing with mission-critical application for the system resources.
We ran a single SQL Server database OLTP workload with a SQL Server buffer pool size of 4 GB. The SQL Server database was running at a Diamond service level for the duration of the test. We started a “noisy neighbor” application at a Silver service level and after some time, we terminated that application. While the test ran, we monitored host IOPS, SQL Server TPM, and read and write response times.
As the test results in the following figure show, PowerMax maintained the performance of the mission-critical application at the Diamond service level while managing a noisy neighbor at a Silver service level for a limited amount of time.
Figure 11. No impact to mission critical application due to noisy neighbor
The SQL Server database at the Diamond Service level achieved read and write response times of 0.17 ms with host IOPS of 231 K and SQL Server TPM of over 2.1 M. As the noisy neighbor application ran, it achieved 1.84 and 1.7 ms read and write response times respectively with SQL Server host IOPS 30 K and TPM of 295 K. Even when it was running there was only marginal impact on the mission-critical application, which continued with 229 K host IOPS and 2.1 M SQL server TPM at the latencies of 0.2 ms for read and write operations. And when the noisy neighbor was terminated, the mission-critical application returned to its original level of operation.
The PowerMax system with SCM drives and other performance features can provide unprecedented levels of IOPS at very low latencies for mission-critical applications. It can maintain that performance level even when some low-priority applications are added to the workload.