Conflating Reliability and Endurance in SSDs

Last week I sat through a presentation by Samsung on the evolution of their DRAM and flash business. One question raised by a journalist in the room related to the reliability of flash drives. The presenter proceeded to talk about the industry and how endurance means flash has a finite lifetime, one which now varies depending on the specific media type and product.

This discussion conflated the idea of reliability and endurance into one. Of course in the flash world, the two are completely separate. Reliability is typically quoted in MTBF (Mean Time Before Failure) or AFR (Annual Failure Rate). Samsung’s 850 PRO quotes MTBF of 2 million hours (or AFR of 0.44), which is in line with all modern hard enterprise hard drives.

So from a reliability perspective, flash drives are as reliable as any of today’s HDDs, however remember that MTBF and AFR are statistical measures of reliability; you could easily buy a drive that fails tomorrow. That’s why we continue to use RAID and erasure coding protection.

What is Endurance?

Endurance is a feature exclusive to SSDs. Writing to solid-state drives results in some degradation of the substrate from which the NAND gates are built. Eventually the NAND cells become unreliable and can’t be used to store/retrieve data. Vendors mitigate this problem using various techniques that optimise the number and distribution of writes across the device. In addition, flash array vendors will write to their SSD devices in a “flash friendly” way to reduce wear overhead.

Flash vendors quote endurance either in terms of the amount of data that can be written to a device over its lifetime (Terabytes Written, or TBW) or in terms of the amount of entire device writes that can be sustained daily over a set period, typically 5 years (Device Writes Per Day or DWPD). Products vary significantly in DWPD capabilities – for example SanDisk Lightning drives vary from 3 to 25 DWPD, depending on the model. This will also be reflected in the price of these devices.

The Architect’s View

So, when looking at flash, remember these devices have an equal (and in some cases higher) reliability than hard drives. The difference is they will eventually fail. However, with many vendors offering 7 or 10 year warranties, concerns about moving to flash because of reliability are rather unfounded.

Comments are always welcome; please read our Comments Policy first. If you have any related links of interest, please feel free to add them as a comment for consideration.