Sign in with
Sign up | Sign in

When Enough is Enough

Can MLC Endure Your Data Load?
By
Brought to you by What's this

Declining endurance levels may not be a problem for owners if their usage levels are low enough to not risk failure within the drive’s service life. How long is that? The answer is getting more tricky. Some vendors still cite a time period, such as two or five years, but expect to see more drives shipping with “media wear-out indicators” integrated into the device. This is essentially a count-down mechanism, and when it runs out, the warranty is up.

Mainstream MLC drives are often built to endure one to two drive writes per day. (A 120 GB SSD writing 120 GB within 24 hours would constitute one drive write.) At such load levels, manufacturers expect the drives to perform reliably within their warranty periods and still have some additional margin of safety. If users start placing higher loads on the drives, say three or four drive writes per day, the warranty may expire early or the drive controller may begin throttling data bandwidth. In effect, this forces the drive to write less data in a given time period in order to preserve the vendor’s warranty assurance.

Different vendors will implement different “dynamic throttling” methods in this pursuit. Enterprise buyers in particular should pay attention to these throttling specifics. In environments where performance is paramount, a shorter warranty coverage period may be justified. Light-use applications may prefer longer service periods.

The cost dynamics of endurance should also be carefully weighed. Consider a 512 GB SSD designed for two drive writes per day with a three-year warranty. That’s 512 (GB) * 2 (drive write) * 365 (days) * 3 (years), or a long-tern data endurance (LDE) of 1.12 PB. In other words, a user will be able to write just over 1,120 TB of data to the NAND media before it reaches its end of life.

One of the ways in which consumer and enterprise SSDs diverge is in the amount of data they are expected to write daily. An enterprise drive may specify support for five or ten drive writes per day while a consumer drive may only allow for 20 GB written per day. A data center that brings in high-performance MLC SSDs with a 20 GB per day write expectation may be in for a very rude awakening when its drives begin to fail prematurely. Such numbers must be run before purchasing. It’s very probable that a business might have to buy six, eight, or even more MLC drives to achieve the same LDE as one enterprise-class drive of the same capacity.

Another related cause of premature SSD failure is write amplification, a phenomenon in which more data is physically written to NAND media than was issued by the host system. This is caused by the inherent nature of SSDs, which require data in a block’s memory pages to be copied out into available space so that the block can be erased and then written to. Several factors contribute to write amplification, but some drives use LSI SandForce controllers containing controller-based compression technology. Again, depending on several factors, this can result in less data being written to NAND than issued by the host. Tom’s Hardware took an in-depth look at this subject just over a year ago.