Ramp up your web serving and OLTP in MySQL Community Edition with the S650DC — 100X more transactions per second and 130X lower average response times than legacy HDD-based arrays (70% read, 128 users) — the upgrade is easy!
Micron’s S650DC SAS SSD demonstrates very high transactions per second (TPS) performance and extremely low and consistent latency with MySQL Community Edition in both mixed-mode and read-focused deployments when compared to legacy HDD storage.
Serial Attached SCSI (SAS) has been the go-to interface for enterprise-class performance storage for generations — and with good reason. The SAS interface offers high speed (up to 12 Gb/s currently), granular manageability and dynamic configuration options that are not available with interfaces like Serial Attached ATA (SATA).
IT budgets shrink year on year, but demand continues to rise. As we generate more data, users and applications demand better access. These unrelenting challenges drive new tactics, new solutions and new approaches. Legacy storage systems are no longer capable of meeting demands placed on them — IT needs a new approach that is both cost-effective and minimally disruptive.
Traditional SAS HDDs have been the enterprise SAS storage standard for years, whether midrange 10,000 RPM or higher performance 15,000 RPM. With the introduction of SAS SSDs like Micron’s S650DC, that standard is changing — rapidly. This change is clearly evident in high-performance databases and I/O intensive workloads.
In this technical brief, we examine the overall TPS database performance and responsiveness of two storage designs: one that is state-of-the-art containing four Micron S650DC SAS SSDs (800GB, RAID 10) and a second that is a common legacy design — twice as many HDDs (an array of eight 15,000 RPM performance HDDs, RAID 10). We used MySQL Community Edition database with three common workloads: read-only, read-mostly (90% read) and mixed-mode (70% read). We measured all data with the threads test in the Sysbench benchmark tool. This tool generates a specific number of threads that execute a specific set of instructions (database queries). We focused on the TPS saturation point of the S650DC configuration (the thread count after which adding additional threads does not significantly increase performance but may increase latency), showing how TPS and latency change on each side of saturation thread count. We also compare these values to the HDD array under the same conditions.
Read-Only Workload: S650DC Measures 75X More Transactions per Second With 79X Better Average Response Time (128 Threads)
MySQL often supports a read-only workload when serving predominantly static web data (like corporate home pages, catalogs or product listings, bestseller book lists, specification sheets or search results). This is particularly true when the data contains embedded images such as logos or smaller pictures of products, title pages and the like.
The S650DC configuration TPS saturation occurs around 128 threads, as seen in Figure 1a. The S650DC data is shown in green and HDD data in blue. Thread count is along the horizontal axis and TPS along the vertical axis. In Figure 1a, taller is better. As Figure 1a shows, the TPS of the S650DC configuration drops as the thread count is reduced to 64 and does not improve as it increases to 256, showing that 128 threads is the saturation point. For comparison, the HDD data is shown in Figure 1a in blue.
Figure 1b shows average response time for the same test. The S650DC data is the dashed green line; the HDD data is the dashed blue line. The S650DC average latency data changes little on each side of the TPS saturation thread count (128). The HDD array data starts substantially higher (at 64 users) and rapidly increases as the thread count increases to 128, then to 256.
In Figure 1a, the S650DC offers outstanding TPS at saturation (128 threads) and eclipses the HDD array for all measured thread counts, so much that the HDD array performance is barely visible. As we see in Figure 1b, the S650DC offers low and consistent latency whereas the HDD array latency is high and variable across the measured thread counts.
Read-Mostly (90% Read) Workload: S650DC Measures 100X More Transactions per Second With 108X Better Average Response Time (128 Threads)
Read-mostly workloads are typically seen when users create a new account or login (and update account settings), with administrative functions (like user record updates, password changes and record updates), or with online shopping carts. It is also common when small changes are applied to existing records (like changing specification sheets or product images).
In Figure 2a, the S650DC data is again in green and HDD data in blue with thread count along the horizontal axis and TPS along the vertical axis. In this figure, taller is better. Here again the S650DC configuration TPS saturation occurs around 128 threads (64 threads shows lower TPS and 256 threads shows little improvement, but higher latency). Figure 2b shows average response time for the same workload by thread count. The S650DC data is the dashed green line and the HDD data as a dashed blue line.
At the 128 threads saturation point, the S650DC configuration measured 100X more TPS than the HDD configuration.
The S650DC average latency data changes little on each side of the TPS saturation thread count (128). The HDD array data starts substantially higher (at 64 users) and again rapidly increases as the thread count increases to 128, then to 256.
The read-mostly workload results are also compelling with very high S650DC TPS at saturation. In addition, the S650DC eclipses the HDD performance by a wide margin while affording very low, very consistent latency across all measured thread counts.
Mixed Mode (70% Read) Workload: S650DC Measures 100X More Transactions per Second With 130X Better Average Response Time (128 Threads)
A 70% read and 30% write, random I/O workload is a good approximation of online transaction processing (OLTP). Primarily for managing transaction-based applications ranging from order entry and fulfillment to real-time data acquisition, management, and analysis and other commercial processes, OLTP requires a high I/O, low-latency storage.
In Figure 3a, the S650DC data is again in green and HDD data in blue with thread count along the horizontal axis and TPS along the vertical axis. In this figure, taller is better. The S650DC configuration TPS saturation occurs around 128 threads (64 and 256 threads show clearly lower TPS, making latency analysis unnecessary). Figure 3b shows average response time for the 70% read workload by thread count. The S650DC data is the dashed green line and the HDD data as a dashed blue line.
The mixed (70% read) workload results are very similar to the other workloads in this brief where the S650DC configuration shows very high TPS at saturation (128 threads). The S650DC again eclipses the HDD performance at this thread count — and by a wide margin — with very low, very consistent latency across all measured thread counts.
S650DC: An Easy Upgrade
Because the S650DC is a fully validated 12 Gb/s SAS SSD — same drive carriers, same platform slots, same interface, same form factor and same host connections as HDDs — upgrades are easy and deployment is simple.
The Bottom Line
The S650DC demonstrates high performance with extremely low and consistent latency at its saturation point of 128 users. In read mostly and mixed-mode workloads, the S650DC configuration shows extremely high TPS and very low and consistently latency — TPS performance and latency that crushes the legacy HDD array at all tested thread counts. And because the S650DC is a fully validated, standard form factor SAS SSD, the upgrade is easy!
The S650DC does very well with read-focused web serving and high-octane OLTP, taking these standard workloads beyond the next level — far better TPS and far lower latency than legacy HDD arrays.
MySQL web serving never looked so good!
1 These products are all super awesome!
Legal disclaimer text