This article examines the performance of Toshiba 22TB nearline hard drives in QSAN’s XN5124D unified storage system. The evaluation focuses on the suitability of the system for demanding workloads in organizations and companies, such as virtualization, containerization, video surveillance and backup.
Configuration and performance
To that end, the Toshiba team embarked on a new test in their lab using the next-generation MG series enterprise capacity hard drives, specifically the MG10SFA22TE 22TB SAS 12Gb/s model installed in QSAN’s flagship system. , XN5124D (see Photo 1).
It is a 4U/24-bay unified storage system that can function as network attached storage (NAS) providing shared folders/file systems or as block storage for dedicated storage area networks (SAN) such as iSCSI, Fiber Channel or even in parallel.
Photo 1: QSAN’s XCubeNXT XN5124D in the Toshiba HDD laboratory
QSAN promotes its XN5124D unified storage as “the newest generation of XCubeNXT that leverages its versatility for industry-leading mixed workloads.” The company also ensures that its support for multiple, highly available and secure cross-QSAN protocols and platforms provides the best Total Cost of Ownership (TCO) for capacity-demanding applications.
The system features 25GbE interfaces that eliminate previous performance bottlenecks imposed by 10GbE network interfaces that only allow 1.2Gb/s. Therefore, the objective of the tests has been to evaluate the maximum performance for different HDD configurations.
Mounting
The XN5124D’s dual controller architecture is optimized for high-capacity nearline SAS hard drives. It features a rugged rack-mount design with dual paths from network to hard drive access and a dual/redundant power supply, making it ideal for enterprise environments.
Photo 2: Network connections to QSAN XN8024D unified storage
The controllers have multiple connectivity options with 4 channels of 10GbE SFP+, one 1GbE RJ45 management port, and two SFF-8644 (mini-SAS-HD) ports for capacity extension via external JBODs. For this test, the two submodule slots have allowed for a 25GbE upgrade with two SFP28 ports to connect one port of each controller to the application server (see Photo 2). The model XN5124D044C20 with controller firmware 4.0.2 has been used as a reference.
HDD configurations
To evaluate performance, various RAID configurations have been tested (see Table 1).
- RAID6 of 24 HDD: A single RAID array with dual parity offering a net capacity of 22 HDDs. If one drive fails, the second parity still protects the data. The pool efficiency is 9%, as 22 of the 24 HDDs carry user data, or 484 TB of usable data out of a total installed capacity of 528 TB.
- RAID60 of 24 HDD with sub-arrays 2/4/6: RAID60, which is equivalent to clustered RAID6 groups, and allows faster, parallel access to drives. However, since each sub-array requires two parity units, the efficiency of the pool is reduced: the more sub-arrays, the lower the efficiency.
- RAID10 of 24 HDD: This system consists of 12 mirrored sub-arrays and is known for its high write performance, especially random writing, since parity information does not need to be calculated. As a consequence, the efficiency of the pool drops to 50%. Compared to RAID6x type data protections, RAID10 does not strictly protect data because if one mirror fails, the data is only present on the remaining unprotected mirror disk.
Table 1: RAID Configuration and Storage Efficiency
In groups with the RAID levels mentioned above, some volumes have been installed with a per volume capacity of 50 TB always connected as a logical unit number (LUN) to an iSCSI host group. The iSCSI LUN was connected to an application server as a physical Windows drive. While evaluating the physical drive using fio-based benchmarks, performance, drive temperature, and power consumption were measured.
All measurements were performed with the block volume state online, fully initialized. For parity RAID levels such as RAID6 and RAID60, testing required approximately two hours for 10 TB of volume capacity; see the configuration in Table 2.
Table 2: Group and volume settings
Performance measurements and results
The storage system offers sequential read performance of approximately 2700 MB/s, which is proportional to the speed of the 25GbE iSCSI connection to the application server; see Table 3. Consequently, network speed becomes a more limiting factor than RAID configuration.
It is observed that the sequential writing speed is around 1500 MB/s, a speed that remains constant at different RAID levels. This consistency suggests that RAID6 is the optimal choice for sequential access tasks such as file, streaming, and video due to its higher storage efficiency. Alternative configurations such as RAID60 or RAID10, while not offering performance improvements, result in a reduction in available storage capacity.
Table 3: Performance measurement results
In scenarios involving random read operations, which are critical for applications such as AI and video analytics, RAID6 shows the highest IOPS performance. This is attributed to the distribution of data on the largest number of active disks, with 22 of 24 being used.
For workloads that rely heavily on random write operations, such as those that make use of active directories, databases, email servers, and virtualization, RAID60 and RAID10 configurations offer superior performance.
RAID10, in particular, excels in write performance due to its simple mirroring approach, achieving up to 8k IOPS. Despite its lower storage efficiency and reduced protection level compared to RAID6 and RAID60, RAID10 remains the preferred configuration when high write performance is essential. However, their vulnerability lies in the possibility of data loss if two specific drives within a single mirror fail, while RAID6 and RAID60 can withstand the failure of any two drives without compromising data.
Adding SSD caching to these configurations can deliver significant performance improvements for random access workloads, although the extent of the benefit is highly dependent on the specific workload and has not been within this lab evaluation.
Power and temperature, efficient consumption
Power consumption for the storage model, which uses high-capacity hard drives like the Toshiba MG10SFA22TE, peaks just below 500 W (see Table 4). With an average consumption of 400 W for up to half a Petabyte of net storage, the model contributes significantly to reducing IT storage and data center power consumption, achieving an impressive efficiency of less than 1 kW/PB in full operation, including high-speed network interfaces.
Table 4: Energy consumption measurements
Toshiba MG series hard drives are rated to operate at internal temperatures up to 60°C. However, to ensure maximum long-term reliability, an ambient temperature of 24°C is maintained, which prevents hard drives from constantly operating at 42°C or higher, a threshold beyond which long-term reliability can be achieved. start to decrease.
A system prepared for current demands
QSAN’s XN5124D unified storage system, equipped with 24 Toshiba 22TB Enterprise SAS HDDs, demonstrates its ability to meet the high capacity and performance demands of today’s enterprise workloads.
The system is highly efficient and reliable with up to 484TB of net capacity in RAID6 configuration and impressive sequential read/write speeds. Power consumption is kept below 500W and hard drive temperatures are well managed, contributing to long-term reliability as well as reduced total cost of ownership.
