This technical brief describes the measured performance advantages of Micron’s 9100 MAX PCIe NVMe SSD compared to a legacy HDD array.
We compared two identical platforms — one with a 1.2TB 9100 MAX RAID 1 array and another with a 1.8TB legacy HDD RAID 10 array. We used Microsoft SQL Server as the test database and standardized OLTP performance metrics to measure the performance and responsiveness of each across a broad number of simultaneous users.
The 9100 MAX configuration showed between 17.7X and 34.9X better performance (measured in orders per minute, or OPM) and 18.1X to 34.5X better average response times that were 4.8X to 38.1X more consistent than the legacy HDD array (with consistency measured as the 99.9th percentile response time).
Microsoft’s SQL Server 2014 is a well-recognized, widely deployed relational database management system (RDBMS), frequently managing transaction-based applications such as order entry and fulfillment and real-time data acquisition/management/analysis (all OLTP-type workloads).
Successful OLTP platforms for modern e-commerce and real-time businesses have to support immediate access to mission-critical data. Order entry/fulfillment and data acquisition/analysis/management are just two examples where delays in data access can be very costly. The database platform must enable fast transaction processing and ultra-low response time. Accelerating transaction processing enables additional transaction execution and management in the same timeframe for more positive outcomes. Reducing the database response time improves RDBMS responsiveness to user requests; users are not idle waiting on storage I/O processes to complete. The first place to look for these improvements is the platform storage system.
OLTP Performance: 17.7X to 34.9X More Orders per Minute With the 9100 MAX Configuration
When compared to SSDs — especially performance-focused, NVMe SSDs like the 9100 MAX — even the fastest HDDs can be painfully slow. This becomes very clear when we compare these two storage platforms using a high-transaction, response time sensitive workload like OLTP.
In this study, we compared two Microsoft SQL Server 2014 platforms, identical in every way with the sole exception of the storage systems. One platform used a pair of 9100 MAX NVMe SSDs (1.2TB each) configured in a RAID 1 array, while the other platform used twelve HDDs (15,000 RPM, 300GB each) configured in a RAID 10 array. These configurations yield similar capacities (1.2TB and 1.8TB, respectively) and represent common performance-focused storage configurations for their respective drive types.
We measured performance for both the 9100 MAX and legacy HDD configurations across a broad number of simultaneous users (8 to 192). Figure 1 shows the results graphically with the number of simultaneous users on the vertical axis and database performance on the horizontal axis. The 9100 MAX configuration results are in green and the HDD results in blue. In Figure 1, a longer bar for any user count is better because it shows greater performance.
Figure 1: Performance by Storage Type
We show the same data numerically in Table 1 with the user count in the far left column and the performance advantage for the 9100 MAX in the far right column.
Table 1: 9100 MAX Calculated Performance
We calculated the performance advantage for the 9100 MAX using the below equation for each user count:
9100 MAX Performance Advantage = 9100 MAX Performance/HDD Performance
For example - looking at a user count of 192 we see:
9100 MAX Performance Difference = 741,309 OPM/41,830 OPM = 17.7X
The 9100 MAX-equipped platform shows significantly better orders per minute than the legacy HDD array
across all user counts, as shown in both Figure 1 and Table 1.
Database Responsiveness: 9100 MAX Has 18.1X to 34.5X Better Average Response Time and Is 4.8X to 38.1X More Consistent
Fast performance is highly desirable in OLTP systems, but responsiveness (and consistency) may be more so in many time-sensitive deployments. We measured both the average and 99.9 th percentile response times (latencies) for the same two storage configurations using the same test metrics.
Figure 2a shows average response time (in milliseconds, along the vertical axis) with user count increases along
the horizontal axis (left to right). Figure 2b shows 99.9th percentile response time (in milliseconds, also along the
vertical axis) with user count increasing along the horizontal axis (left to right). In Figures 2a and 2b, lower is
better for any user count.
Figure 2a: Average Response Time by Storage Type
Figure 2b: 99.9th Percentile Response Time by Storage Type
The 9100 MAX configuration measured 18.1X to 34.5X better average response time with 4.8X to 38.1X better
consistency than the legacy HDD array (exact values depend on user count).
Note, too, that the 9100 MAX configuration shows flatter (left to right, across increasing user counts) average
and 99th percentile value. At the application level, this means the 9100 MAX-based platform responds more
quickly and more consistently as application loads increase.
The Bottom Line
Delayed database response times can be very costly, whether in access to mission-critical data, order entry/fulfillment or data acquisition/analysis/management applications. Time can literally mean money.
Legacy HDD-based storage is no longer sufficient to meet the incredible demands OLTP places on Microsoft SQL Server 2014 platforms. High-performance OLTP demands high-performance storage like the 9100 MAX SSD.