Accelerate Database Performance
with All-Flash Storage
For many database applications—from high-frequency trading to real-time risk management—microseconds translate to opportunities created or opportunities lost.
Since the speed with which I/O can be completed is critical to database performance, enterprises are increasingly turning to all-flash storage to minimize I/O latency and accelerate transactional performance.
Choosing the right all-flash storage translates directly to the bottom line with big improvements in processing time, less server hardware needed, lower database licensing costs, and significant savings in floor space and energy usage. (For more on this check out Flash Delivers Storage Performance for an Impatient World in this issue.)
Driving Maximum Database Performance
For years, database administrators (DBAs) have carefully planned and controlled data storage to squeeze out the best possible performance from relatively slow disks, with best practices that included running busy log files on fast, mirrored drives to deliver the best available performance. DBAs also favored dedicated, over-provisioned infrastructure to avoid “noisy neighbor” contention and unpredictable response times.
All-flash arrays substitute solid-state disks (SSDs) for slow, mechanical disks. The results are dramatic for every element of performance. A flash SSD delivers hundreds of times more IOPS than a hard disk drive (HDD) and slashes access latency from milliseconds to microseconds.
But there are important architectural differences that make various all-flash arrays more or less well suited for high-performance database workloads. A clean, optimized data path is essential to take full advantage of flash latencies measured in microseconds. It is critical that storage services don’t get in the way of fast, consistent performance.
The new NetApp® EF560® all-flash array is ideally suited to deliver maximum all-flash performance for business-critical databases. This new platform delivers state-of-the-art performance results for both absolute and consistent performance. With up to 650K IOPS in a compact 2U form factor and average response times of 800 microseconds or less, the EF560 shows the importance of storage system design. Furthermore, the EF560 includes the proven reliability and availability developed over eight generations of the SANtricity operating system.
The EF560 is designed to satisfy the best practices of the high-performance DBA. Storage CPUs are dedicated to fast read and write activity while host resources are used for compression. Nothing in the code path gets in the way of processing I/O, so the array not only delivers latency measured in microseconds under load, it delivers the same latency on day 1,001 as it did on day 1—with no unexpected spikes.
For DBAs, this dedicated performance is consistent with the drive to realize consistent predictable results. We’ve characterized the performance and price/performance of the EF560 with a variety of internal benchmarks (IOPS and throughput) plus the SPC-1 benchmark, and the results are exceptional.
First, let’s look at how the EF560 performs on a read workload—the type that is most often reported for all-flash arrays. The EF560 delivers 650,000 sustained I/O operations per second at 800 microsecond latency from just 2U of rack space. That’s a 62.5% bump over the EF550. If we take 500 microseconds as a performance threshold, the EF560 can still deliver 628,000 IOPS.
We’ve taken the characterization of EF560 IOPS performance a step further to help better predict performance for the workload you intend to run. (The system under test was an EF560 with 48 SSDs, RAID 5, and 8KB block size.)
Table 1) EF560 IOPS performance under various read/write workloads.
Source: NetApp, 2015
SPC-1 Benchmark Performance
To provide additional validation, we have published results for the EF560 running the Storage Performance Council’s SPC-1 benchmark, which does a good job of simulating the demands of a high-performance database with a high percentage of writes as well as reads. (See sidebar About the SPC-1 Benchmark for details.) In this audited test, the EF560 delivers 245,011.76 SPC-1 IOPS™ with an average response time (ART, measured at 100% load) of 0.93 milliseconds or 930 microseconds. If we consider 500 microseconds as a threshold, the EF560 delivers 196,008.41 SPC-1 IOPS at 0.53 milliseconds (530 microseconds) response time. The typical customer has a requirement for 80-120K IOPS—well within the capabilities of the EF560.
Figure 1) EF560 SPC-1 response-time curve.
The EF560 offers the lowest SPC-1 Least Response Time, or LRT (measured at 10% load) of any configuration on the SPC-1 “Top Ten” lists at 0.18 milliseconds (180 microseconds). SPC-1 LRT represents the minimum response time possible from a storage system. View the EF560 SPC-1 executive summary or the full report.
A primary aspect of the SPC-1 benchmark is that it provides price performance in terms of $/SPC-1 IOPS for each configuration tested. The EF560 comes in at $0.54 per SPC-1 IOPS, that puts the EF560 at #2 overall on the “Top Ten” list for SPC-1 Price-Performance™. If you look at results with average response time under 1 millisecond, the EF560 is the #1 array.
For a deep-dive on the EF560 SPC-1 results, check out this blog post from Dimitris Krekoukias on RecoveryMonkey.org.
Throughput or bandwidth is another important aspect of flash array performance. While it’s talked about less frequently than IOPS, throughput is a measure of how well an array can read or write sequential—rather than random—data. All databases have an important sequential component (think log files). Analytics workloads and in-memory databases can see huge benefits from fast streaming for data ingest.
This is another area where the EF560 stands out. Where other flash competitors brag about 4GB/sec or 8GB/sec read throughput, the EF560 can sustain 12GB/sec read throughput and more than 6GB/sec write throughput.
EF560 Performance Enhancements
The EF560 takes advantage of four new performance enhancements:
EF560 Manageability and Availability Enhancements
EF560 manageability and availability improvements include:
The EF560 features dual-active controllers housed in a 2U shelf and configured with up to 120 SSDs (currently in 400GB, 800GB, and 1.6TB capacities). A single array delivers up to 192TB of raw capacity. SSDs are configured in volumes using either Dynamic Disk Pools (DDP) or RAID 0, 1, 5, 6, and 10.
Figure 2) EF560 specifications.
Source: NetApp, 2015
Choosing a Data Protection Method
One of our goals for the EF560 was to fully characterize performance and operations to make it easy for you to choose the best configuration for your needs. With DDP growing rapidly in popularity—35% of EF-Series deployments already use it—this includes guidance on when to choose DDP versus another RAID level.
To choose the right data protection, look at your requirements and imagine each of the possible protection schemes as performance zones as shown in Figure 2. Then choose the level of protection that delivers the performance and capacity you need.
Figure 3) EF560 performance using RAID5, DDP, and RAID 10.
Source: NetApp, 2015
Keep in mind that this figure is based on a workload with 8KB block size, 75% reads and 25% writes, and a latency ceiling of 0.6 milliseconds (600 microseconds). Increasing the percentage of writes moves the dividing lines down for each technology.
The DBA’s Choice for Database Performance
If you’re looking for fast, consistent performance from an all-flash array, look carefully at the architecture. The EF560 all-flash array features a dedicated performance architecture that is ideal for business-critical high-performance databases. With a legacy approaching 1 million installed storage systems and a reliability of more than five 9s, The EF560 has an architecture that performance DBAs can rely on.
Visit Tech OnTap in the NetApp Community to subscribe today.