Mixing HDDs in RAID

This article was written by 2 storage specialists from Toshiba Electronics Europe GmbH:

Irina Chan, senior specialist product marketing management storage products, and

Rainer W. Kaese, senior manager business development storage products

Mixing Hard Disk Drives in RAID Systems
RAID System Reliability: Do Native and Emulated Block Size HDDs Mix?

RAID systems combine multiple disks into a single logical unit to improve the performance and reliability of data storage. However, like any other piece of electronic equipment, they can be subject to failures, errors, or degradation. Thay must therefore be properly maintained and optimized to ensure they function reliably. This includes replacement and upgrading of HDDs when necessary.

But HDDs now come in 2 different types of block size: native and emulated. This raises questions about whether these different types can be mixed within the same RAID system, and what effect this might have on system performance and reliability.

The Toshiba HDD Laboratory recently evaluated the flexibility of 512n and 512e replacements in RAID systems.

The test regime sought to address several questions:
– Are 512n and 512e block sizes fully interchangeable in RAID systems?
– Where no replacement candidate of the initial capacity is available or at hand, will a drive of larger capacity do the job just as well?
– What if a replacement drive or the original manufacturer is not available? Will a similar drive from another HDD manufacturer work as replacement drive?

Test conditions
The tests first considered native and emulated HDDS in NAS and enterprise storage. The Toshiba NAS HDD N300 was used as the representative 4TB version with 512n block size, and the enterprise model of the MG-Series MG08ADA400E as the 512e block size. These are the most common 4TB drives and this capacity is often used in smaller RAID sets of typically 2 or 4 bays.

A failing disk was simulated by simply unplugging it under full power and operation (hot-removal), and waiting for 10mn while checking the performance and function of the RAID array in this failed drive mode. The replacement drives were then inserted, also under full power, with time allowed for RAID rebuilding without any power cycle.

Assessments were based on HDD performance in 2 common 4-bay desktop-NAS models (the QNAP 4-bay Desktop NAS TS-464-4G and the Synology 4-bay Desktop NAS DS-420+) and several popular RAID controllers, with the HDDs installed in the front bay of the server chassis (the Microchip Adaptec SmartRAID 3204, Microchip Adaptec SmartRAID 3154, Broadcom MegaRAID SAS 9460 and its successor the Broadcom: MegaRAID SAS 9560).

RAID technology is also used for DAS, which connects a RAID-based storage box to the host via dedicated USB or Firewire connections, so tests were also conducted to confirm the replacement flexibility in a 4-bay RAID from Icy-Box and a SafeTANK GR5640-SBA31+ from Raidon.

End users from datacenters and large-scale enterprises to small/medium businesses and private users may also use older legacy systems, even beyond the original manufacturer’s warranty and support time frames. In terms of RAID controllers in particular, there is still a significant installation base of historic 3ware models, so this evaluation also took into consideration the replacement of failing drives on older RAID sets of 512n drives with 512e replacement drives, as well as in a RAID system with a more than 10-year-old 3ware 9750-4i controller.

Encouraging results
The results revealed that, for all configurations, failing 512n drives can indeed be replaced by 512e models without restrictions. The reverse is also true: replacing a failing drive in a 512e RAID with a 512n drive works in all cases. So, mixing 512n and 512e is possible for all RAID configurations, although re-build strategies differ between the RAID engines tested.

Further, a failed drive can be replaced by a model with higher capacity, but the additional capacity of the replacement drives remains unused. These tests also indicated that HDDs of the same interface (SATA), same block size (512B, regardless of emulated or native) and same- or larger capacity can also be mixed, regardless of the manufacturer. However, previous evaluations by Toshiba (1) have shown that there are limits to mixing. Drives with 512B and 4k byte block size cannot be mixed or used as replacement drives, and neither can drives with different interface technology (SATA and SAS).

Interestingly, the RAID rebuild time of the legacy controller was in the same range as the other RAID systems. This indicates that the RAID rebuild performance depends on the speed of the drive, rather than on the performance of the RAID controller.

RAID recommendations
The ability to mix native and emulated HDDs in virtually any RAID configuration is good news for system users and operators, but there are other important aspects to consider which will aid system optimization and avoid problems during disk replacement.

Firstly, it must be remembered that RAID technology is not a backup, it simply helps you to continue working in case of a storage media failure. If the RAID controller/box fails or other storage bugs occur, all your data will be lost. Backing up the data on your RAID/NAS onto a portable component (the Toshiba Canvio 2.5” external USB-connected HDD, for example) for storage in a separate, external location will mitigate against data loss.

HDD failure cannot be identified by simply observing RAID function and performance. It’s therefore important to check the indication LEDs on RAID boxes and NAS systems regularly. Thanks to their short point-to-point connectivity, RAID boxes are usually placed close to the host computer and are easily observed. Due to their network connectivity, NAS boxes can be located anywhere; often in cabinets, racks or cellars, where a red-blinking indicator LED is easily overlooked. Either login to the control dashboard of your system on a regular basis and check the status, or configure the system so that an error message is sent out in the event of failure.

A replacement disk should be acquired in advance of any failure event. Ideally it should be the same model as the original RAID set but, as our latest testing demonstrates, there is a lot of flexibility. Just make sure it’s a SATA model, it has the same or higher capacity as the original, and it is not a 4kn native block size.

Once the replacement drive is available, swap the failing drive and check the system to see whether automatic rebuild has started. If not, start it manually in the respective menus. Although you can continue working with the data, it is best to leave the RAID system idle as this will shorten the rebuild time. Keeping rebuild times as short as possible is important because your data is rather vulnerable during re-build as there is no redundancy. Another drive failure while re-building will therefore kill your data.

Failed drives should never be inserted back into the RAID system. If a drive has failed once, it will fail again. Even though it may appear that all is going well following re-insertion, many RAID systems get confused if they recognize a drive that has been part of a RAID configuration before by finding known metadata on the drive. They may try to re-insert it into the RAID set, or assume it’s part of a different RAID. The results can be catastrophic.

And finally, whether to use all four drives in RAID-5 (single parity) or RAID-10 (striped and mirrored) configuration, or whether to buy just 2 drives of double the capacity and run in RAID-1 (simple mirror) mainly depends on capacity and speed requirements. Make sure to check your requirements before making the final choice of RAID system.

In summary
Evaluation indicates that there is good forwards and backwards compatibility between native and evaluated HDDs of equivalent block size, even in older legacy RAID systems. Further, there is no discernible effect on RAID system performance in mixing native and evaluated HDDS within a single system. This offers system owners and operators greater flexibility in sourcing replacement drives. By following best practice during disk replacement and rebuild, the integrity of system data can be assured and system performance optimized, regardless of HDD choice.

This article necessarily contains only a summary of the results from Toshiba’s latest native vs. evaluated HDD test program. Full details are available in a new whitepaper, which is free to download here.

References:
(1) Enterprise HDD Legacy Systems
How New HDDs Interact with Old RAID Controllers