Kingston recently released an enterprise class solid state drive designed for high permanent loads. Now many journalists are actively testing the novelty and producing interesting materials. We want to share with Habr one of the detailed reviews of the Kingston DC500R, which readers will enjoy testing. The original is on the website. and published in English. For your convenience, we have translated the material into Russian and place it under the cut. Enjoy reading!
Storage devices based on 3D TLC NAND flash memory technology. Available in 480 GB, 960 GB, 1,92 TB, and 3,84 TB capacities, this provides an additional choice for businesses looking to save money or those who just don't need high capacity drives. This review is based on the 3,48 TB variant, which claims sequential read and write speeds of 555 MB/s and 520 MB/s, respectively, and read and write speeds of 4 KB blocks under constant workloads of 98 and 000 inputs. -outputs per second (IOPS) respectively. As part of this product family, Kingston also offers the DC28M, which is optimized for mixed-use applications.
Specifications Kingston DC500R
Performance
Test
A system was used to test enterprise-grade SSDs with real-world applications. , and for synthetic testing - . The ThinkSystem SR850 is an optimized quad-core platform that delivers far more processing power than you need for high performance local storage testing. For synthetic benchmarks where CPU capabilities are not as important, a more traditional two-CPU server was used. In both cases, we were hoping to get local storage performance in line with manufacturer specifications.
Lenovo ThinkSystem SR850
- 4 Intel Platinum 8160 processors (2,1 GHz, 24 cores)
- 16 x 4GB DDR2666 ECC DRAM 32MHz
- 2 x RAID 930-8i 12 Gb/s adapters
- 8 NVMe drives
- VMware ESXI 6.5 Software
Dell PowerEdge R740xd
- 2 Intel Gold 6130 processors (2,1 GHz, 16 cores)
- 4x 4GB DDR2666 ECC DRAM 16MHz
- PERC 730 12 Gb/s RAID Adapter 2 GB Buffer
- Embedded NVMe Adapter
- OS Ubuntu-16.04.3-desktop-amd64
Test Information
provides ample opportunities for testing storage devices in an environment close to real conditions. The laboratory includes various servers, network devices, power systems and other network infrastructure. This allows our employees to create realistic conditions for accurately assessing equipment performance.
Information about the environment and protocols is included in the reviews so that IT professionals and those responsible for purchasing storage systems can evaluate the conditions under which the corresponding results were achieved. The manufacturers of the equipment under test do not pay for or control the review.
Application workload analysis
To correctly evaluate the performance parameters of enterprise storage devices, it is important to model the infrastructure and application workloads that correspond to real environments. Therefore, to evaluate Samsung 883 DCT SSDs, we measured и using TCP-C workload emulation. In this case, for applications, each drive will handle 2 to 4 identically configured virtual machines.
SQL Server Performance
Each SQL Server virtual machine is configured with two virtual disks: a 100 GB boot disk and a 500 GB disk to house the database and log files. In terms of system resources, each virtual machine was equipped with 16 virtual processors, 64 GB of DRAM, and an LSI Logic SAS SCSI controller. Previously, we used Sysbench workloads to test I/O speed and drive capacity efficiency. SQL tests, in turn, help to estimate latency.
As part of testing, SQL Server 2014 is deployed on guest virtual machines running Windows Server 2012 R2. Workloads are created using Quest's Benchmark Factory for Databases software. StorageReview uses the current version of Benchmark C (TPC-C) from the Transaction Processing Performance Council. This real-time transaction processing performance benchmark simulates the processes of complex application environments. TPC-C testing can better identify the strengths and weaknesses of storage infrastructure in database environments than artificial performance testing. In our testing, each instance of the SQL Server VM was running a 333 GB (1500 scale) SQL Server database. Transaction processing performance and latency were measured with a load of 15000 virtual users.
SQL Server test configuration (for each VM):
• Windows Server 2012 R2
• Disk space: 600 GB allocated, 500 GB in use
• SQL Server 2014
- Database size: 1 500 scale
— Number of virtual clients: 15
- RAM buffer: 48 GB
• Test duration: 3 hours
– 2,5 hours – preliminary stage
- 30 minutes - direct testing
The Kingston DC500R was only slightly behind the Samsung 883 DCT in SQL Server transaction processing performance, with an overall throughput of 6290,6 transactions per second (TPS).
An even better way to evaluate SQL Server performance than TPS is to measure latency. Here, both drives - Samsung 860 DCT and Kingston DC500R - showed the same time: 26,5 ms.
Performance when using Sysbench
The following test used the database . OLTP performance was evaluated using the SysBench utility. This measures the average TPS and latency, as well as the average latency under the worst-case scenario.
Each virtual machine I used three virtual disks: a boot disk with a capacity of about 92 GB, a disk with a pre-installed database of about 447 GB, and a disk with a test database with a capacity of 270 GB. In terms of system resources, each virtual machine was equipped with 16 virtual processors, 60 GB of DRAM, and a SAS SCSI controller from LSI Logic.
Sysbench test configuration (for each VM):
• CentOS 6.3 64-bit
• Percona XtraDB 5.5.30-rel30.1
— Number of database tables: 100
- Database size: 10
— Number of database threads: 32
- RAM buffer: 24 GB
• Test duration: 3 hours
- 2 hours - preliminary stage, 32 streams
— 1 hour — direct testing, 32 streams
Based on Sysbench transaction processing performance, the DC500R outperformed the competition with a performance of 1680,47 transactions per second.
In terms of average latency, the DC500R also ranked last with 76,2ms.
Finally, after testing the latency under the worst-case scenario (99th percentile), the DC500R was again at the bottom of the list with a result of 134,9 ms.
VDBench workload analysis
When testing storage devices, application testing is preferred over synthetic tests. However, although their results do not correspond to real conditions, synthetic tests, due to the repeatability of tasks, are convenient for establishing baselines and comparing competing solutions. These tests offer a wide range of profiles, from "four corners" tests and generic database migration tests to tracking captures from various VDI environments. All of these use a single vdBench workload generator with a script engine to automate and collect results on a large cluster of compute tests. This makes it possible to use the same workload across a wide variety of storage devices, including all-flash arrays and single drives. As part of our testing, we completely filled the drives with data and then partitioned them into 25% of the original capacity to simulate application workloads and evaluate the behavior of the drive. This approach differs from fully entropy tests, in which the entire disk is used at once under constant loads. For this reason, the following results reflect more stable write speeds.
Profiles:
• 4 KB random read: read only, 128 threads, I/O rate from 0 to 120%
• Random write 4 KB: write only, 64 threads, I/O rate from 0 to 120%
• 64 KB sequential read: read only, 128 threads, I/O rate from 0 to 120%
• 64 KB sequential write: write only, 64 threads, I/O rate from 0 to 120%
• Synthetic databases: SQL and Oracle
• VDI Copy (Full Copy and Linked Copy)
In the first VDBench workload test (4K Random Read), the Kingston DC500R delivered impressive results, with latency within 1ms up to 80 IOPS and a top speed of 000 IOPS with a latency of 80ms.
All tested drives performed almost identically in the second test (4K Random Write), with slightly over 63 IOPS at 000ms latency.
Moving on to sequential loads, we first looked at 64K reads. In this case, the Kingston drive kept latency within a millisecond until reaching 5200 IOPS (325 MB/s). The maximum score of 7183 IOPS (449 MB/s) with a latency of 2,22 ms brought this drive to the second place in the overall standings.
When testing sequential write operations, the Kingston device outperformed all competitors, keeping the latency below 1 ms up to a speed of 5700 IOPS (356 MB / s). The maximum speed was 6291 IOPS (395 MB / s) with a delay of 2,51 ms.
After that, we moved on to SQL tasks, where the Kingston DC500R was the only device that exceeded the latency limit of a millisecond in all three tests. In the first case, the drive showed a maximum speed of 26411 IOPS with a latency of 1,2 ms.
In the SQL 90-10 test, the Kingston drive finished last with a top speed of 27339 IOPS with a latency of 1,17ms.
The same thing happened in the SQL 80-20 test. The Kingston device in this case showed a maximum speed of 29576 IOPS with a latency of 1,08 ms.
Oracle's workload test results once again placed the DC500R in last place, but the device still showed sub-millisecond latency in two tests. In the first case, the maximum speed of the Kingston drive was 29098 IOPS with a latency of 1,18 ms.
In the second test (Oracle 90-10), the DC500R achieved 24555 IOPS with a latency of 894,3 µs.
In the third test (Oracle 80-20), the maximum speed of the device from Kingston was 26401 IOPS with a latency level of 831,9 µs.
Then we moved on to copying VDI - making full and linked copies. In testing loading a full copy of the VDI, the Kingston drive again failed to beat the competition. Below 1ms latency, the device was able to support up to about 12000 IOPS, and topped out at 16203 IOPS with a latency of 2,14ms.
In the Initial Login test of the VDI copy, the Kingston device performed better, finishing (by a slight margin) in second place. The drive maintained latency within a millisecond until reaching a speed of 11000 IOPS, and the maximum speed was 13652 IOPS with a delay of 2,18 ms.
Also by a small margin, the Kingston drive came in second in Monday Login testing for a full copy of VDI. The Seagate Nytro 1351 drive showed slightly better top speeds, but the Kingston device generally showed lower latency levels throughout the test. The maximum speed of the DC500R was 11897 IOPS with a latency of 1,31 ms.
In testing downloading linked VDI copies, the Kingston came in last place. The latency went beyond 1ms already at speeds less than 6000 IOPS. The maximum speed of the DC500R was 7861 IOPS with a latency of 2,03 ms.
However, in the Initial Login test, the drive again took second place: the delay went beyond the limits of a millisecond only near the peak performance, which eventually amounted to 7950 IOPS with a latency of 1,001 ms.
In the last test of the linked copy of VDI - Monday Login - the drive also showed the second result: the maximum speed at the level of 9205 IOPS with a latency of 1,72 ms. The delay went beyond a millisecond when the speed reached 6400 IOPS.
Conclusion
The DC500R is Kingston's latest solid state drive for enterprise users. The DC500R comes in a 2,5" form factor. Capacity options are available ranging from 480 GB to 3,84 TB. The drive is based on 3D TLC NAND flash memory technology and combines a long resource and a high level of performance. For a disk with a capacity of 3,48 TB, sequential read and write speeds of 555 and 520 MB / s, respectively, read and write speeds at constant loads of 98000 and 28000 IOPS, respectively, as well as a resource amount of 3504 TBW, are declared.
To evaluate the performance of the Kingston DC500R, we compared it to other popular SATA SSDs, including Samsung drives и , as well as storage . Kington DC500R was able to keep up with the competition, and in some cases even surpass them. When testing application workloads, the Kingston DC500R performed well in processing SQL tasks, finishing second overall in TPS (6291,8 TPS) and latency (26,5 ms). In Sysbench testing with more write-intensive workloads, the DC500R came in at the bottom of the list with a performance score of 1680,5 TPS with an average latency of 76,2ms and a worst-case latency of 134,9ms.
The Kingston DC4R achieved 500 IOPS and 80209ms read latency in 1,59KB random read and write tests, and 63000 IOPS and 2ms write latency. In 64KB read and write tests, the DC500R achieved speeds of 7183 IOPS (449 MB/s) at 2,22 ms latency and 6291 IOPS (395 MB/s) at 2,51 ms latency, respectively. Under the conditions of synthetic tests using SQL and Oracle databases and increased requirements for write speed, the performance of the DC500R left much to be desired. When it comes to SQL workloads, the Kingston DC500R came in last place in all three tests and was the only drive to exceed millisecond latency. However, in testing Oracle, the picture turned out to be much better. In two of the three tests, the drive kept latency below 1ms, which earned it a second-place finish. The Kingston DC500R delivered a decent level of performance when tested with VDI copies, both full and linked.
Generally — a quality device in its class, which deserves closer attention. As much as we love high-performance technologies (NVMe and similar), SATA drives remain the preferred solution for tasks where reliability is critical, such as loading a server or storage controller. These drives are also a cost-effective server storage solution for situations where value for money is critical. In addition, they offer all of the total cost of ownership advantages that distinguish solid state drives from hard disk drives (HDDs). The DC500R's performance puts it at the top of many of our tests against other hard drives worth considering. The DC500R is an excellent SATA drive for scenarios that require high capacity, reliable, high performance drives in a wide range of capacities.
Models of the DC500 series are available to order from official Kingston distributors.
For testing and validation questions, please contact the Kingston Technology representative office in Russia at the e-mail address [email protected]
For more information about products refer to the company's website.
Source: habr.com