jules - Fotolia

Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

Does TLC flash offer enough resilience for enterprises?

The suitability of triple-level cell (TLC) flash storage for enterprise use depends on whether applications are heavily read- or write-optimized. Expert Marc Staimer explains the ramifications.

It depends on how TLC flash storage -- appearing as 3D NAND flash drives -- is used for reads and writes, how much it is overprovisioned, the ultimate effectiveness of the flash storage controller's wear-leveling and error-correcting code (ECC).

One well-known secret to all-flash NAND storage is that reads are free, but writes cost. Reads have an effect on flash storage wear, but writes significantly wear flash storage. This is because the data written on flash NAND program-erase (P/E) blocks cannot be changed in place. To write to a P/E block, it must not have anything written on it before the write occurs, or it must be erased before anything can be written to it.

It should be noted that flash storage drives are capacity overprovisioned. Capacity overprovisioning allows the flash storage to nearly always have an available P/E block unwritten or erased for the writes so an application does not have to wait for P/E blocks to be erased. It also enables the flash storage to replace worn-out P/E blocks or blocks with unrecoverable bit errors, so rated capacity is not lost.

In general, TLC flash has a much lower write wear-life than multi-level cell (MLC) and single-level cell (SLC) flash.

Capacity overprovisioning is not part of flash storage-rated capacity claims. For example, a 2 TB flash SSD may have 10% or 20% overprovisioning. An SSD with 10% overprovisioning will have a rated capacity of 1.92 TB, while one at 20% will have a rated capacity of 1.64 TB.

Demartek LLC founder and president Dennis Martin talks about developments in solid-state storage with Rich Castagna, editorial director of TechTarget's Storage Media Group.

The amount of overprovisioning generally determines the use case for that flash storage. High levels of capacity overprovisioning make flash storage suitable for frequent writes. Lower levels make flash storage more optimized for reads.

So how does TLC flash storage resilience factor into this? In general, TLC flash has a much lower write wear-life than multi-level cell (MLC) and single-level cell (SLC) flash. TLC is typically rated at less than 1,000 P/E cycles, MLC ranges from 10K to 30K P/E cycles and SLC is commonly above 100,000 cycles.

Flash storage controller software that provides clever wear-leveling across all the blocks plus extensive ECC software can increase the potential P/E cycles by as much as 3x. The amount of overprovisioning can also help increase P/E cycles.

Read-optimized TLC flash storage, advertised as read optimized or prioritized, is very likely qualified for enterprise use. Read-optimized typically means ≤ 10% writes and ≥ 90% reads. This makes TLC flash storage suitable for enterprise applications that are heavily read-oriented, such as the second tier of a storage system for passive data, active archiving, backup and recovery, analytics, business intelligence, content distribution networks, video streaming and so on. Write-optimized TLC flash storage is far less likely to be qualified for enterprise use and is not recommended.

Next Steps

Why triple-level cell flash storage makes sense

When to expect TLC flash in enterprise storage

How TLC entered the MLC vs. SLC fray


Dig Deeper on Solid-state storage

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

What TLC flash features are most appealing for enterprise use?
I'm still trying to figure out the use case for needing the speed of flash for something that people only read.
@Sharon Fisher -- It really depends on how frequently that data is read. If it is a "small" amount of data which is read frequently by many users (example: Wikipedia, Google search index, Netflix movie catalog), then read intensive SSDs make lots of sense. It all depends on the IOPs / GB of data you need in order to access the data. More users accessing the same data means more IOPs/GB are needed. The newest HDDs actually are declining significantly in IOPs / GB, as the IOPs stay basically around <300 IOPs, but the capacities grow and grow. This means HDD performance is declining in terms of IOPs/GB, and in many cases the decline opens up even more opportunity for read intensive SSDs. (Disclaimer - I am an Intel Corporation employee within the SSD product division. This position is my own, and is not the voice of Intel.)
Interesting nuance, thanks!