When an IT professional starts looking into solid state drives (SSDs) they quickly learn that flash is very different from magnetic disk drives. Flash employs a much more complex write and erase process than traditional hard disk drive (HDD) technology does, a process that impacts performance, reliability and the device’s lifespan (flash eventually wears out). To address potential concerns, vendors have historically sold different types of flash, with the more expensive SLC (Single-Level Cell) being used in more critical environments. But with advances in controller technologies, is the more economical MLC (Multi-Level Cell) good enough for the enterprise?
SLC or MLC
As indicated above there are actually different types of NAND flash used in storage products. SLC NAND supports two logical states enabling each cell to store one bit of data. MLC NAND, and its “enterprise” cousin eMLC, have a capacity of up to four bits per cell. This increased density gives MLC a lower cost per bit stored but also creates a higher probability of bit errors. SLC has historically been faster, more reliable and exhibited greater endurance than MLC or eMLC.
The flash controller does run error correction code (ECC) but it’s only effective to a point, after which device reliability starts to decrease. The flash substrate also degrades a little bit each time it’s erased and re-written, exacerbating the bit error problem, something that’s even more of an issue for MLC flash than it is for SLC. This means that in a typical flash implementation the type of flash chosen has definitely been important.
These factors helped make SLC the original flash technology of choice for enterprise-grade products, such as flash drives, PCIe cards and flash modules used in storage arrays. MLC was originally only for consumer uses like thumb drives and flash memory cards for cameras since these products typically saw less write activity than did flash-based storage products for the data center.
MLC in the Enterprise
As error correction technologies have advanced (due in part to increased flash controller processing power) MLC NAND reliability and endurance have both improved significantly. In the end, if a device can record and reproduce data accurately, it doesn’t matter if that device is experiencing fewer bit errors or the flash controller is just working harder ‘under the covers’ correcting those errors. This has resulted in MLC NAND making its way into enterprise-grade products, bringing down their effective costs and opening up more use cases for flash in general.
Improving Reliability and Endurance
Advancements in flash software have continued to increase the controller’s ability to correct bit errors, which in turn has extended the useful life of MLC-based products. Some of these technologies use parity-like schemes to improve data integrity as each block traverses the data path inside the flash controller. Others include ‘tuning’ the read processes to best fit the condition of the flash substrate (since flash changes as it ages).
In addition to improving the controller’s ability to correct errors and accurately read data, vendors have worked on the write and erase processes as well to reduce their impact on endurance. Since the flash degrades a little as it’s used, some vendors have developed processes that are more ‘gentle’ to the flash substrate, helping reduce its degradation over time.
A Tipping Point
Due to these innovations in error correction and new ways to write and erase data, some vendors are selling MLC flash products that actually have better endurance and reliability than their own SLC products did even a year ago. The result is more flash availability and lower costs per GB. In fact, in the past year or so the industry has reached a ‘tipping point’ of sorts regarding the effective difference between SLC and MLC (or eMLC) NAND. MLC flash is becoming the primary technology in enterprise flash products.
Vendors are now designing their flash devices to provide a level of reliability and performance for a specified length of time and often using different technologies to get there. Some are using SLC flash or even DRAM as an input buffer to reduce the wear on the MLC flash that’s providing the bulk of the device’s capacity. And many are not even mentioning which type of flash they’re using.
Flash products are being built to support a minimum number of bytes written throughout their life cycles, called Total Bytes Written (TBW). This can be extrapolated into the number of times a given product can be filled with data, completely erased and filled again before it ‘wears out’. This number is typically expressed as ‘fills per day’ with a 3-5 year lifespan being common in enterprise storage products.
For users the message is clear. Determine the endurance needed by calculating the amount of data likely to be written to flash during its projected lifetime and then shop for products that specify that level of endurance. In order to make this kind of comparison, it’s incumbent upon the flash vendors to provide TBW and reliability specs that the user has confidence in. For the record, not all vendors do this today.
Storage Swiss Take
The flash industry is maturing. Thanks to advancements in software, which come much more quickly and at less development cost than do those in hardware, the reliability and endurance of lower cost flash technologies like MLC, have greatly improved. This in turn has shifted the decision factors from the type of flash used to what the performance and endurance are for each vendor’s products. Users really don’t need to know what specific flash technology is at work, they just want their manufacturer to clearly state what the product will do and then deliver on that promise.
This focus is also appropriate given the fact that NAND flash is reaching the end of its technological road. Sub 20nm flash (“1x” nm) is out now but the industry is looking to other technologies to drive solid state storage into the future. Focusing on the performance and endurance a device delivers and not what it’s made of will make it easier for vendors to integrate these new solid state technologies into the storage products we’re using and keep delivering the improvements that technology consumers have come to expect.