While we talk about solid-state drives (SSDs) as an example that most people are familiar with in storage systems, the same controller- and system-related myths also apply to USB flash disks, eMMC modules, SD cards, or any other form of managed NAND device. It’s time to separate fact from fiction.
1. The latest flash generation is always the best.
New flash-memory generations hit the market with one mission: Squeeze more gigabytes of information into the costly silicon. As a result, the price per gigabyte decreases with newer generations of flash. However, this usually means fewer electrons are associated with one bit of information. Hence, the quality isn’t getting better per se. It’s becoming more complex to manage.
While this isn’t critical for the consumer market, where quality and reliability are second to price, the fact that the latest NAND flashes can be used in highly demanding applications is due to the close collaboration of flash and controller vendors.
2. Performance doesn’t depend on capacity.
Flash-array write and read times are slower than the system performance. To reach a certain performance, controllers need to interleave read and write operations, meaning access different flash arrays at the same time. Especially for write performance, it’s necessary to connect several flashes and hence provide higher capacity. On the other hand, newer flashes increase the per-die capacity so that an SSD based on a newer flash technology may be slower at the same overall capacity.
3. When you compare the performance of different SSDs, bigger is always better.
Datasheets usually promote attractive performance and often refer to the burst speed of the interface. However, looking at the system, often the performance bottleneck of the system is either in the controller or limited by the flash technology. Most significantly, while the fresh-out-of-the-box performance is great, performance, especially random write speed, deteriorates as soon as the drive is filled.
Many systems are optimized to look outstanding new with popular benchmark programs. However, depending on your usage, the real experience can differ significantly. Using cheap flashes and cheap controllers never has a performance benefit in the long run.
4. Performance is constant over lifetime.
The more you fill up your SSD, especially with random data, the more administration and management that must be done in the background by the controller. The amount of intelligence that a controller can apply to interleave background management tasks with your actual read or write activity can be game-changing. Cheaper controllers with fewer resources in terms of internal memory or hardware accelerators may do a poor job that either results in a shorter lifetime of the system or significant performance drops.
5. NAND-based storage solutions are independent of the customer use case.
The specific use case of a customer application heavily influences the performance and lifetime of a storage device. Therefore, it’s important to know as much as possible about your specific application. With this knowledge, you can save a lot of money and avoid customer complaints.
SLC, MLC, and TLC NAND flash memories.
6. Consumer NAND flash controllers are always cheaper.
Cheap controllers have less features and hardware support to adequately reflect the production price. However, non-optimal wear leveling, error handling, or flash management will indisputably lead to a shorter lifetime, and your SSD may fail. Consequently, you lose all of your valuable data and your system is left non-functional. Since most of your system cost is determined by the flash and not the controller, investing in a quality controller certainly pays off and guarantees a lower total cost of ownership (TCO).
7. Data is gone when you erase it.
Due to the nature of the NAND flash, it’s safe to assume that your data is still on the NAND flash even after you’ve deleted the files from your storage solution. The actual erase process is only performed during certain management procedures like garbage collection. Certain controllers offer secure erase capabilities, but the real nature of the erase must be understood if it’s important for the end application.
8. SATA is disappearing with the rise in PCIe sales.
While PCI Express (PCIe) has dominated in many consumer markets such as gaming and laptops, the rise and demanding nature of industrial data storage has left the SATA interface with a strong footing in industrial markets. With demand rising for SSD storage, SATA SSDs are still highly sought after and will address markets where PCIe performance isn’t required or power budget is limited. New SATA flash-memory controllers can be used with the latest 3DFlash memory, providing a price-competitive solution for applications not driven solely by performance.
9. The more static data, the longer the lifetime.
As NAND Flashes have a limited number of write/erase cycles, one can assume that if a drive has a large amount of static data, it’s likely to have a longer life expectancy. This isn’t the case. In fact, two problems arise through static data. First, retention can become an issue if the data is never rewritten. Second, wear leveling often can’t be carried out on static data. High-quality NAND controllers can feature sophisticated algorithms to wear-level static data to blocks that have more usage and, therefore, prolong the lifetime of a drive.
10. The function of wear leveling is carried out equally.
Wear leveling is a vast subject and can be very basic for consumer applications. Alternatively, it can be very complex when the end application demands longevity and reliability. It’s important to dig deep into wear-leveling features with flash controller vendors to understand the true robustness of a system. In short, if your data is critical, or has significant value, it’s preferable to look into controllers designed for demanding requirements.
11. High-speed interfaces are only advantageous.
It’s not only about the sheer amount of memory connected to the controller that influences performance and usability. It’s equally as important that you manage the heat created by a device running at 1 GB/s or even higher speeds. Even active water cooling is under discussion for high-end NVMe solutions, which comes at a cost. Ultimately, one should not assume that high-speed interfaces are delivered free—they incur further cost to run.
Damien Col is Technical Marketing Manager at Hyperstone.