Toshiba's A19nm MLC NAND has opened up new opportunities for density and performance increases. Toshiba has responded by releasing a dual-pronged update to their SATA 3.2 (6GB/s) enterprise SSD offerings. The HK3R2 eSSD addresses read-centric workloads with a new higher capacity option (up to 960GB), and enhanced write performance. The HK3R2 is designed for read-centric workloads with one DWPD (Drive Write Per Day) of endurance for five years, satisfying requirements for read caching, video streaming workloads, and general datacenter storage applications.
The HK3E2 is a value-endurance model that utilizes A19nm NAND to deliver up to three DWPD for five years. It addresses mainstream enterprise applications, such as email servers, web servers, database, and indexing workloads. Both SSDs feature similar components to reduce qualification requirements, and similar performance specifications, with the exception of the HK3E2's increase to 30,000 random write IOPS.
Endurance measurements are with a 4KiB random write workload, and are worst-case scenarios due to the nature of the test workload. Mixed random workloads and sequential workloads produce less wear, thus extending endurance beyond rated specifications. The HK3R2 at 960GB offers incredible resilience for a read-centric SSD with 1.76 Petabytes of total endurance. This ranks above the majority of other contenders in the read-centric segment.
The HK3R2 is the second-generation of the Toshiba HK3R Enterprise SSD we recently evaluated. The Toshiba HK3R2 Series comes in capacities of 120, 240, 480, and 960GB in the 2.5" form factor with a 7mm z-height. The HK3R2 utilizes the proprietary Toshiba TC358790XBG controller, and features a 4k random read speed of 75,000 IOPS. 4k random write performance varies by capacity, but tops out at 14,000 IOPS for the 960GB model. The HK3R2 provides a 524 MB/s sequential read rate, but sequential write speed also varies by capacity, topping out at 419 MB/s.
The HK3R2 leverages several key technologies to protect data, and we will take a closer look at these approaches on the following page. In short, Toshiba's proprietary QSBC (Quadruple Swing-By Code) error correction scheme provides an UBER rating of one per 10^17.
Toshiba also has two different levels of power loss protection, and the HK3R2 utilizes both PLP (Power Loss Protection) and PFM (Power Fail Management). Two large capacitors allow the HK3R2 to persist all data from its 1GB of DDR3 DRAM to the NAND in the event of a host power-fail situation. The SSD also features end-to-end data protection, and is designed to be incredibly power efficient with a 16,600 IOPS-per-watt rating.
Two key differentiators in the extremely competitive value-centric SSD market are endurance and data protection techniques. We dive deeper into Toshiba's approach on the following page.
HK3R2 Power Loss Protection and QSBC
One of the most basic requirements for any storage device is the ability to identify and repair data errors. Denser and more cost-effective NAND technologies, such as TLC NAND and smaller NAND lithographies, come with increased bit error rates. Traditional SSD error correction implementations utilized BCH ECC to combat data errors, but higher bit error rates have led to LDPC (Low Density Parity Check) development. LDPC features two levels of error correction, hard-decision and soft-decision, which enhances correction capability, and reduces latency. Real-time LDPC requires more compute power than BCH ECC, and a tradeoff is that decoding circuits are larger and consume more power as a result.
LDPC has several advantages to previous error correction approaches, but Toshiba has taken a divergent path with their proprietary QSBC (Quadruple Swing-By Code) error correction technology. Flash controller manufacturers are developing custom LDPC algorithms, and each will feature different characteristics, so a direct comparison of the varying approaches is not entirely possible. Many of the details on custom LDPC codes, along with QSBC, are closely guarded intellectual property.
The graphic above indicates the error correction capability of three typical implementations. Typical Mini-Sum LDPC offers a 2x increase over BCH, but Toshiba's QSBC outpaces Mini-Sum LDPC by 8x. Toshiba has a long history with manufacturing NAND and overcoming errors, so it is not surprising that they already have an advanced proprietary solution integrated into their products.
Toshiba also utilizes End-to-End error detection and correction to protect every step along the data path, and NAND flash.
PLP and PFM
SSDs utilize cache layers to boost performance, but the cache is typically volatile DRAM that is susceptible to power loss. Power loss results in data loss, if there is not a mechanism to commit all data to the NAND when the SSD loses power. Drive management tables (LBA tables) and the firmware can also be compromised during power loss. Protecting data during power loss requires capacitors, and firmware optimizations.
The HK3R2 features both PFM (Power Fail Management) and PLP (Power Loss Protection). The Toshiba PLP implementation features a detection circuit that detects power loss and notifies the firmware, which then activates the Power Line Switch to shift power to the onboard capacitors. The SSD immediately initiates a rapid shutdown, and flushes cached data to the NAND.
PFM works in tandem with PLP to protect the internal management tables. Redundant copies of the management tables are stored on different physical NAND pages, and updates to the tables are alternated between the copies. The system simply restores the redundant copy during power-up if a table is corrupted. PFM also ensures that all data in the cache is flushed every two seconds during idle periods, and that new data is always written to a different NAND page. Toshiba validates every Toshiba SSD with PLP and PFM through 30,000 power cycles during their internal RDT process.
Toshiba HK3R2 Internals and Specifications
Toshiba HK3R2 Internals
The Toshiba HK3R2 THNSNJ960PCS3 comes in a 2.5" form factor with a slim 7mm z-height in a feather-light alloy case. The large indentation on the bottom of the case raises thermal pads up to mate with the components mounted on the PCB.
The HK3R2 utilizes small thermal pads that are strategically placed to wick away heat from the components, and transfer it to the case. The PCB is secured into the case with four additional fasteners.
The HK3R2 features eight Toshiba A19nm MLC Toggle NAND packages, and two large capacitors. These capacitors flush any data in transit down to NAND in the event of a power failure. 1GB of Micron DDR3 800MHz DRAM serves as the cache for the SSD. There are no surface mounted components on the bottom of the PCB, but additional thermal pads pull heat from the rear of the packages.
The proprietary Toshiba TC358790XBG controller powers the HK3R2. This controller was also utilized in the HK3R series, providing it with a track record of reliability. The Toshiba designed TC358790XBG controller powers the HK3R2.
Toshiba HK3R2 Specifications
Test System and Methodology
Our approach to storage testing targets long-term performance with a high level of granularity. Many testing methods record peak and average measurements during the test period. These average values give a basic understanding of performance, but fall short in providing the clearest view possible of I/O QoS (Quality of Service).
While under load, all storage solutions deliver variable levels of performance. 'Average' results do little to indicate performance variability experienced during actual deployment. The degree of variability is especially pertinent, as many applications can hang or lag as they wait for I/O requests to complete. While this fluctuation is normal, the degree of variability is what separates enterprise storage solutions from typical client-side hardware.
Providing ongoing measurements from our workloads with one-second reporting intervals illustrates product differentiation in relation to I/O QoS. Scatter charts give readers a basic understanding of I/O latency distribution without directly observing numerous graphs. This testing methodology illustrates performance variability, and includes average measurements during the measurement window.
IOPS data that ignores latency is useless. Consistent latency is the goal of every storage solution, and measurements such as Maximum Latency only illuminate the single longest I/O received during testing. This can be misleading, as a single 'outlying I/O' can skew the view of an otherwise superb solution. Standard Deviation measurements consider latency distribution, but do not always effectively illustrate I/O distribution with enough granularity to provide a clear picture of system performance. We utilize high-granularity I/O latency charts to illuminate performance during our test runs, and our IOPS to Latency measurements highlight IOPS performance at specific latency points.
Our testing regimen follows SNIA principles to ensure consistent, repeatable testing, and utilizes multithreaded workloads found in typical production environments. We measure power consumption during precondition runs. This provides measurements in time-based fashion, with results every second, to illuminate the behavior of power consumption in steady state conditions. We also present IOPS-to-watts measurements to highlight efficiency.
This evaluation features SSDs of varying capacity, and overprovisioning variations require consideration during performance and power consumption analyses. The Toshiba HK3R2 sample has 960GB of user-addressable capacity, the Micron M500DC features 800GB of capacity, and the Samsung 845DC EVO and Intel DC S3500 models are 480GB. The SSDs are tested over their full LBA range to highlight performance at maximum utilization. The first page of results will provide the 'key' to understanding and interpreting our test methodology.
Benchmarks - 4k Random Read/Write
4k Random Read/Write
We precondition the 960GB Toshiba HK3R2 THNSNJ960PCSZ for 9,000 seconds, or two and a half hours, receiving performance reports every second. We plot this data to illustrate the drive's descent into steady state.
This dual-axis chart consists of 18,000 data points, with the IOPS on the left, and the latency on the right. The blue dots signify IOPS, and the grey dots are latency measurements during the test. We place latency data in a logarithmic scale to bring it into comparison range. The lines through the data scatter are the average during the test. This type of testing presents standard deviation and maximum/minimum I/O in a visual manner.
Note that the IOPS and latency figures are nearly mirror images of each other. This illustrates high-granularity testing can give our readers a good feel for latency distribution by viewing IOPS at one-second intervals. This should be in mind when viewing our test results below. This downward slope of performance only occurs during the first few hours of use, and we present precondition results only to confirm steady state convergence.
Each level tested includes 300 data points (five minutes of one second reports) to illustrate performance variability. The line for each OIO depth represents the average speed reported during the five-minute interval. In some charts, we include a smaller embedded chart that lists performance at the highest load for easy interpretation.
4k random speed measurements are an important metric when comparing drive performance, as the hardest type of file access for any storage solution to master is small-file random. 4k random performance is a heavily marketed figure, and is one of the most sought-after performance specifications.
The Toshiba HK3R2 comes out swinging with an average of 85,042 IOPS at 256, second only to the Samsung 845DC EVO, which delivers 85,155 IOPS at 256 OIO (Outstanding I/O). Matching the 845DC EVO in random read speed is impressive, and providing significantly more endurance tailors the Toshiba well for more strenuous activities. The Micron M500DC averages 56,259 IOPS, and the Intel DC S3500 averages 57,769 IOPS.
Our Latency v IOPS charts compare the amount of performance attained from each solution at specific latency measurements. Many applications have specific latency requirements. These charts present relevant metrics in an easy-to-read manner for readers who are familiar with their application requirements.
The HK3R2 joins the 845DC EVO with impressive latency, while delivering tremendous speed. The HK3R2 provides 85,000 IOPS at .1ms (similar to the 845DC EVO), the DC S3700 provides 57,000 IOPS, and the M500 DC provides 56,000 IOPS.
Garbage collection routines are more pronounced in heavy write workloads, leading to performance variability.
The HK3R2 delivers a healthy increase in random write speed along with improved consistency in comparison to its predecessor; it averages 21,583 IOPS at 256 OIO. The Micron M500DC leads the test pool in 4k random write workloads with 39,089 IOPS at 256 OIO.
The Micron M500DC has the lowest overall latency in the write workload, with the HK3R2 taking second place.
Our write percentage testing illustrates the varying performance of each solution with mixed workloads. The 100% column to the right is a pure 4k write workload, and 0% represents a pure 4k read workload.
The mixed workload performance exhibits the tight competition among many competitors in this class. The HK3R2 is very competitive in this test, and features the highest peak speeds, albeit with more variability. The Micron M500DC leads in all but the pure random read workload, and the gap widens from 80-100%.
We record power consumption measurements during our precondition run. We calculate the stated average results after the device has settled into steady state during the last five minutes of the test.
User-addressable capacity should be taken into consideration when viewing power metrics. The HK3R2 has the highest addressable capacity of the test pool, yet still manages to provide impressively low power draw. The HK3R2 averages 3.57 watts, the 845DC EVO averages 3.55 watts, the M500DC averages 4.09 watts, and the DC S3500 averages 3.8 watts during the measurement window.
IOPS-to-watts measurements are generated from data recorded during our precondition run, and the stated average is from the last five minutes of the test.
The HK3R2's IOPS-per-watt is listed at 16,600 IOPS, which is likely measured with read activity. Our measurements focus on write workloads. The HK3R2 averages 6,390 IOPS-per-watt, the 845DC EVO averages 3,973 IOPS-per-watt, and the M500DC takes the lead with 9,545 IOPS-per-watt due to its outstanding write performance. The DC S3500 averages 3,125 IOPS-per-watt.
Benchmarks - 8k Random Read/Write
8k Random Read/Write
Many server workloads rely heavily upon 8k performance, and we include this as a standard with each evaluation. Many of our server workloads also test 8k performance with various mixed read/write distributions.
The average 8K random read speed of the Toshiba HK3R2 is 50,560 IOPS at 256 OIO, the Samsung 845DC EVO measures 52,731 IOPS, the Micron M500DC measures 48,034 IOPS, and the Intel DC S3500 measures 44,444 IOPS.
The HK3R2 places second in the latency testing.
The HK3R2 averages 10,357 IOPS, the M500DC leads with an average of 23,852 IOPS, the 845DC EVO comes in third with 7,112 IOPS, and the DC S3500 averages 6,937 IOPS.
The HK3R2 leads the DC S3500 and the 845DC EVO, but trails the M500DC by a large margin.
The test field separates into two clear groups as we mix in more write activity, with the HK3R2 and M500DC coming out on top.
Power consumption for the HK3R2 averages 3.68 watts; the 845DC EVO averages 3.54 watts, the M500DC averages 4.77 watts, and the DC S3500 averages 3.75 watts.
The HK3R2 averages 2,683 IOPS-per-watt, the M500DC leads the efficiency test with 4,877 IOPS-per-watt, the 845DC EVO averages 1,994 IOPS-per-watt, and the DC S3500 averages 1,584 IOPS-per-watt.
Benchmarks - 128k Sequential Read/Write
128k Sequential Read/Write
128k sequential speed reflects the maximum sequential throughput of the SSD, and is indicative of performance in OLAP, batch processing, streaming, content delivery applications, and backup scenarios. The Toshiba HK3R2 averages 521 MB/s, and the Samsung 845DC EVO squeaks by with an average of 526 MB/s at 256 OIO. The Micron M500DC averages 417 MB/s, while the Intel DC S3500 delivers an average of 441 MB/s.
The 845DC EVO and the HK3R2 place very close in latency metrics during the measurement window.
Sequential write performance is important in tasks such as caching, replication, HPC, and database logging. The Toshiba HK3R2 takes a big lead in this test with an average of 488 MB/s. The 845DC EVO trails with 427 MB/s, the M500DC averages 388 MB/s, and the DC S3500 averages 424 MB/s.
The HK3R2 separates itself from the pack with a very consistent latency profile.
The HK3R2 manages to dethrone the M500DC in mixed sequential performance, which is a tall order considering the gulf between the two leaders.
The HK3R2 continues to impress with sequential workload performance as it averages 3.66 watts, the 845 DC EVO averages of 4.14 watts, the M500DC averages 5.24 watts, and the DC S3500 averages 4.83 watts.
The Toshiba HK3R2 leads the efficiency chart easily with 133 MB/s per watt; the 845DC EVO averages 99 MB/s per watt, the M500DC averages 73 MB/s per watt, and the DC S3500 averages 87 MB/s per watt.
Benchmarks - Database/OLTP and Web Server
This test consists of Database and On-Line Transaction Processing (OLTP) workloads. OLTP is the processing of transactions such as credit cards and high frequency trading in the financial sector. Databases are the bread and butter of many enterprise deployments. These demanding 8k random workloads with a 66 percent read and 33 percent write distribution bring even the best solutions down to earth.
The HK3R2 averages 22,525 IOPS, but suffers much of the same variability as the M500DC, whose average of 21,133 IOPS takes second place. The Samsung 845DC EVO has a very consistent average of 19,678 IOPS at 256 OIO, and the Intel DC S3500 averages 19,400 IOPS at 256 OIO.
The HK3R2 makes a solid gain in latency vs IOPS performance in comparison to its predecessor, and takes the lead in this test.
The HK3R2 averages 3.30 watts, the 845DC EVO averages 3.55 watts, the M500DC averages 2.76 watts, and the DC S3500 averages 3.76 watts.
The chart gets muddy with several SSDs in close competition. HK3R2 averages 7,062 IOPS-per-watt, the 845DC EVO averages 5,465 IOPS-per-watt, the M500DC averages 7,676 IOPS-per-watt, and the DC S3500 averages 4,040 IOPS-per-watt.
The Web Server workload is read-only with a wide range of file sizes. Web servers are responsible for generating content users view over the Internet, much like the very page you are reading. The speed of the underlying storage system has a massive impact on the speed and responsiveness of the server hosting the website.
The HK3R2 leans on its impressive random read speed to take the lead in this test with an average of 28,796 IOPS at 256 OIO; the 845DC EVO averages 28,054 IOPS, and the M500DC averages 18,650 IOPS, falling to the DC S3500 average of 23,664 IOPS.
The HK3R2 pulls away from the 845DC EVO in this workload.
The HK3R2 averages 3.62 watts, the 845DC EVO averages 3.55 watts, the M500DC averages 3.21 watts, and the DC S3500 requires 3.8 watts during the fileserver workload.
The HK3R2 averages 1,303 IOPS-per-watt, the 845DC EVO scores 914 IOPS-per-watt, and the M500DC scores 1,965 IOPS-per-watt, in comparison to 749 IOPS for the DC S3500.
Benchmarks - Email Server
The email server workload is a demanding 8K test with a 50% read and 50% write distribution. This application is indicative of the performance in heavy write workloads.
The HK3R2 averages 18,507 IOPS, the 845DC EVO averages 14,041 IOPS, the Micron M500DC averages 15,403 IOPS, and the Intel DC S3500 averages 13,121 IOPS at 256 OIO.
The HK3R2 continues to take the lead in heavy mixed workloads.
The HK3R2 averages 3.56 watts, the 845DC EVO averages 3.54 watts, the M500DC averages 2.66 watts, and the DC S3500 averages 1.75 watts.
The HK3R2 averages 7,062 IOPS-per-watt, the 845DC EVO averages 3,815 IOPS-per-watt, the M500DC averages 5,806 IOPS-per-watt, and the DC S3500 scores 6,705 IOPS-per-watt.
Toshiba is the only NAND manufacturer with ingrained HDD manufacturing, providing them with an extensive portfolio of datacenter storage products that run the gamut from HDDs up to the fastest of 12Gb/s SAS SSDs. The SATA segment's growing popularity stems from vast compatibility with existing infrastructure, and its lower cost structure compared to SAS SSDs.
Toshiba's HK3R2 and HK3E2 are competitive products that feature a suite of proprietary enhancements to boost endurance and resilience to data errors. QSBC ECC provides a higher level of error correction capability in comparison to most standard approaches, and end-to-end data protection identifies and corrects data errors in flight. A two-tier power loss scheme ensures data integrity during unintentional power loss, and protects the internal management tables and firmware.
The HK3R2 displayed incredible 4k random read speed that rivals the 845DC EVO. The random write speed easily beats the 845DC EVO, and it offers three times the endurance as well, but the HK3R2 comes in second to the M500DC in heavy write workloads. The M500DC fares extremely well in random write testing, but doesn't match the HK3R2 in random read speed. The M500DC also features a lower amount of user-addressable capacity, which boosts performance, but can affect price-per-GB metrics. The HK3R2 emerges as a balanced SSD, well suited for a wide variety of workloads and use-cases, and that translated over to our server workload testing.
The HK3R2 leveraged its impressive random read speed to run away with the lead in webserver testing, and provided leading performance in OLTP tests. The heavy mixed read/write workload in the email server test provides a clear view of how the HK3R2's balanced characteristics provide enhanced performance in the mixed workloads required for smooth application performance. The HK3R2's IOPS-to-latency measurements consistently provided the best performance profile in server workloads.
The HK3R2 also has a clear advantage in sequential activity, where it provides best-in-class performance in mixed workloads. It also led the field in sequential write performance, and placed a very close second in sequential read workloads. Surprisingly, the HK3R2 even ran away with the lead in power metrics during sequential testing. Higher performance usually equates to more power draw, but the HK3R2 provided a much lower power draw than competitors with sequential workloads, even though it has the highest performance and the highest addressable capacity in the test pool. Power consumption metrics across the board revealed the HK3R2 features low power draw, and excellent efficiency.
Toshiba's move to A19nm NAND fields a well-rounded SSD, well suited for mixed random workloads, and sequential read/write workloads. A differentiator is the HK3R2's higher endurance, which is higher than the 845DC EVO (3x), and the DC S3500. The HK3R2 narrowly falls below the M500DC's endurance, with 1.76 v 1.9 PB of endurance. The M500DC features a lower one per 10E15 rating, in comparison to one per 10E17 for the HK3R.
The Toshiba HK3R2 is an impressive entrant into the read-centric class of enterprise SSDs that offers balanced random performance, leading sequential performance, terrific mixed workload performance, and low power consumption metrics. The Toshiba HK3R2 earns the TweakTown Editor's Choice Award.
|Quality, Design, Build and Warranty||94%|
|Power Consumption and Efficiency||96%|
The Bottom Line: Toshiba's HK3R series offers incredible performance in random workloads, and it leads in server workload performance. It also features class-leading sequential performance characteristics, along with impressive low power consumption and great efficiency.
PRICING: You can find products similar to this one for sale below.
United States: Find other tech and computer products like this over at Amazon.com
United Kingdom: Find other tech and computer products like this over at Amazon.co.uk
Australia: Find other tech and computer products like this over at Amazon.com.au
Canada: Find other tech and computer products like this over at Amazon.ca
Deutschland: Finde andere Technik- und Computerprodukte wie dieses auf Amazon.de