Solution Brief The Impact of SSD Selection on SQL Server Performance Understanding the differences in NVMe and SATA SSD throughput 2018, Cloud Evolutions Data gathered by Cloud Evolutions. All product names are the trademarks of their respective owners. 1/5
Relational database management systems (RDBMS) are widely used in enterprise applications such as business analytics, decision support systems, and fraud detection. Optimizing throughput for these systems requires high performing SQL databases to ensure rapid consumption and processing of data for faster time to insight and the ability to make decisions quickly. This is particularly true in financial services where risk assessments of transactions must happen instantaneously to stop fraud before it happens. Toshiba CD5 Series Data Center NVMe SSD Server and storage choices directly impact the ability to deliver the required level of performance for decision support systems. Solid state drives (SSDs) can significantly enhance the ability to consume large amounts of data rapidly, enabling performance to make decisions faster. For SQL databases, NVMe SSDs offer high IOPS and bandwidth, as well as faster response times, which can directly impact speed of processing. Finding the ideal balance of footprint, cost, and performance with NVMe SSDs While many SQL databases exhibit OLTP-type of workloads, which are essentially small block random reads and writes that require high IOPS, some often overlooked decision support system workloads need high sequential throughput. In these situations, large amounts of data must be read or loaded to the database sequentially for processing or for backup purposes. The Toshiba Memory CD5 Series Data Center NVMe SSD is designed for scale-out, data center environments. It supports capacities from 960GB to 7,680GB in a U.2 form factor and has a 9-14W power envelope. It delivers 500,000/35,000 random read/write IOPS and 3,140/1,980MB/s sequential read/write performance. Optimized for low latency and consistent performance under read-intensive workloads, it is a great fit for hyperscale/cloud and Open Compute Project architectures. Today, SATA SSDs are most commonly used in the data center. However, because data in SQL server databases is often from disparate sources, SATA SSDs can struggle to deliver the necessary processing speeds. Cloud Evolutions set out to test the different SATA and NVMe SSD options to quantify limitations and evaluate performance. The questions studied included: How can you improve throughput performance and reduce data transfer times without increasing your storage footprint? What drive throughput performance threshold, can sufficiently reduce backup time and minimize impact in the production environment? Would the performance benefit of NVMe SSDs justify the additional cost? Toshiba CD5 vs. Intel DC S4500 600% throughput performance gain 7x latency improvement 6:1 reduction in footprint Similar cost between enterprise SATA and CD5 Series NVMe SSD 2018, Cloud Evolutions Data gathered by Cloud Evolutions. All product names are the trademarks of their respective owners. 2/5
AMD EPYC Server Processor Family AMD EPYC server processors provide breakthrough processing power balanced with the industry-leading memory and I/O capacity needed to substantially eliminate performance bottlenecks. AMD EPYC powers the data center with 32 cores, 64 threads, 8 memory channels with up to 2 TB of memory per socket, and 128 PCIe-3 lanes coupled with the industry s first hardwareembedded x86 server security solution. With the flexibility to choose from 8 to 32 cores, EPYC enables you to deploy the right hardware platforms to meet your workload needs from virtualized infrastructure to largescale big-data and analytics platforms and legacy line-of-business applications. EPYC, with a balanced set of resources, offers an unprecedented opportunity to rightsize data center infrastructures for enhanced performance and cost efficiency. Cloud Evolutions evaluation approach SATA SSDs are the most commonly used SSDs in the data center. We selected the popular Intel S4500 DC SATA SSD for testing. The new Toshiba Memory CD5 Series Data Center NVMe SSD was also evaluated as a similar cost alternative to SATA. Cloud Evolutions tested both of these on a Dell EMC R7425 server with an AMD EPYC 32 core processor (with 64 threads, 8 memory channels, and 128 lanes of PCIe Gen 3) running MS SQL Server 2016. The Microsoft Performance Monitor (perfmon) was used to capture all data elements that measure throughput related performance, such as disk reads/sec (read IOPS), disk writes/sec (write IOPS), disk sec/reads (read latency) and disk sec/ write (write latency). Key findings - Toshiba CD5 performance, latency and footprint advantage The Toshiba CD5 measured just under 3GB/s throughput, outperforming the Intel S4500 DC by nearly 6X. Extrapolating these results to the data center, you would need 6 SATA SSDs to achieve the same throughput performance of one Toshiba CD5 NVMe SSD. Therefore, SATA SSDs require an increased storage footprint as well as greater power usage. We also measured latency as a component of performance. The Toshiba CD5 demonstrated an average latency of less than 3 milliseconds, compared to a 21ms latency for the Intel drive. This shows a 7X improvement in response time with the Toshiba CD5, which shows significant benefits over SATA, with 6:1 smaller footprint. SQL Server proven performance advantage With the need for high throughput performance to optimize SQL Server databases, tests clearly demonstrate that the Toshiba CD5 NVMe SSDs deliver a measurable advantage. With 6X the throughput and 7X lower latency, the Toshiba CD5 NVMe SSD offers an economical alternative to SATA SSDs in terms of performance, power and storage footprint given the price similarity between the two drives tested. The Toshiba Memory CD5 offers system administrators of MS SQL Server databases and RDBMS great value through shortened data load/backup times, smaller storage footprints, and reduced energy costs. 2018, Cloud Evolutions Data gathered by Cloud Evolutions. All product names are the trademarks of their respective owners. 3/5
APPENDIX A: TECHNICAL SPECIFICATIONS Test Results Speed (MB/s) [Higher is better] Toshiba CD5 NVMe SSD Intel s4500 DC SSD 501 2,976 Latency (ms) [Lower is better] Toshiba CD5 NVMe SSD Intel s4500 DC SSD 2.916 21.171 Server Specifications Server Name Server Vendor Model Operating System Memory (GB) Processor Type and Count CPU Count RAID Controller R7425 DELL EMC Windows Server 16 512 2 AMD EPYC 48 PERC H740p Drive Specifications Manufacturer Name Drive Model Interface User Capacity (GB) Sequential Read (128KB, QD32) No. of Drives Mean Time to Failure Form Factor TOSHIBA CD5 PCIe Gen 3, NVMe 1.3 960 Up to 3140 MB/s 1 2.5M Hours 2.5-inch 15mmH Intel DC S4500 SATA 3.0 6Gb/s 960 Up to 500MB/s 1 2M Hours 2.5-inch 7mmH Application Environment SQL Server version SQL Server Edition Operating System Type of Workload SQL Server Tool Test Database 2016 Developer Windows Server 2016 Datacenter OLTP Management Studio 200GB 2018, Cloud Evolutions Data gathered by Cloud Evolutions. All product names are the trademarks of their respective owners. 4/5
Monitoring Tools Performance Monitor Specifications Operating system Windows Server 2016 Datacenter Performance counters Processor time (CPU usage) Average Disk Bytes/Transfer (Throughput) Disk Sec/Transfer (Latency) Performance Monitor Performance Monitor is a System Monitoring tool which is available by default on all Microsoft Windows machines. It monitors system performance counters such as RAM usage, CPU usage, response time, data block size, and many more, which are useful in analyzing the performance of a machine and determining the cause for a performance problem. This performance counters data is even helpful to identify system bottlenecks and to fine-tune the system and application performance. To analyze the performance of any application or our local machine itself, this tool was used to collect real-time data and to store data logs for the specific period we requested. The performance counters that we used for collecting data to analyze the performance of the Toshiba CD5 SSDs was: Average Disk sec/transfer (Latency), Average Disk Bytes/sec(Throughput), Disk Transfer/sec (IOPS), Processor Time (CPU usage). NVM Express and NVMe are trademarks of NVM Express, Inc. Dell is a trademark of Dell Inc. Microsoft and Windows, Windows Server are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries. PCI Express and PCIe are trademarks or registered trademarks of PCI-SIG. All other company names, product names, and service names mentioned herein may be trademarks of their respective companies. Information in this document including products, availability, specifications, technical/application data and contacts are current and believed accurate on the date of publication, but is subject to change without prior notice. 2018, Cloud Evolutions Data gathered by Cloud Evolutions. All product names are the trademarks of their respective owners. 5/5