Is it time for all-flash datacenters?

Server Room

April 28, 2023

It’s commonplace to say that flash memory, a non-volatile form of data storage, is the obvious replacement for hard disks when it comes to high speed and low latency, but is this already a reality in the datacenter world?

For Pure Storage experts, there are many reasons to adopt flash memories fully into datacenter installations, given the dizzying growth of unstructured data globally as well as sustainability issues. In addition, the price per bit of NAND flash memory is falling at a much higher rate than hard disk drives, so 2023 should be the year we see the cost barrier fall for flash memory to gain more space in datacenters.

In addition to the price of flash memory itself, another cost category will soon make disks the least economical option as a storage resource in datacenters – that of power. With power rates around the world rising, it is natural to consider how much each type of data storage in datacenters consumes.

On the hard disk drive side, their moving parts rely on motors that consume a lot of power and generate a lot of heat, and consequently, more power needs to be devoted to cooling. Flash memory, on the other hand, has no moving parts, so it consumes less power for operation and cooling. In addition, as the density of flash memory continues to increase, its energy efficiency will soon surpass that of hard disks, helping to reduce the costs associated with power consumption. One more point in favour of flash memories.

Another economic issue in this puzzle has to do with licensing costs. The purchase process for old datacenter storage systems was complex and had hidden values. Besides the price of the hard disk arrays, over time the need for complementary resources and tools that were not included in the basic configuration arose, which generated extra costs, not to mention those of support and maintenance contracts. In the case of flash memories, the main suppliers work with hardware and software packages, for example, for deduplication and compression. According to Pure Storage, it is enough to compare the BOM (Bill of Materials, list of items that compose a product) of an all flash storage system and a traditional storage system to prove that the former is composed of only two or three items. This translates into lower acquisition and maintenance costs.

Another unquestionable advantage of flash memories is performance. It is well known that the speed of hard disks is very low, even in undemanding scenarios. Flash memory has no performance problems, with the added benefit that the speed at which data is read remains predictable even when the drives are loaded with data, unlike hard disks, where the speed at which data is accessed depends on its location in the storage unit.

The third advantage of flash memory over hard disks is reliability. The failure frequency of flash memory is significantly lower. The direct economic consequence is that more reliable devices need less maintenance and fewer replacements or, in other words, less expense.

Types of flash memory

There are four types of flash memory known by the acronyms SLC, MLC, TLC and QLC. The letter ‘C’, for cell, represents a floating gate transistor that can assume two states (or voltage levels) associated with 0 or 1. The more cells, the higher the density of stored data and, naturally, the higher the cost.

SLC or Single Level Cell means that each cell contains a single bit. It is the fastest, most reliable, accurate and, of course, most expensive type of flash memory technology. MLC or Multi Level Cell usually means two bits per cell. It is the most widely used type of flash storage technology. TLC or Triple Level Cell means three bits per cell. TLC technology has made all-flash storage more economically viable. QLC or Quad Level Cell means four bits per cell and has the lowest cost per gigabit. Pure Storage experts caution that if a vendor recommends against using QLC because it is unreliable, too slow or the cost difference between QLC and TLC is not significant, beware. It is likely that the legacy architecture it provides is not compatible with newer, QLC technology.

Summarizing all these elements, it is safe to say that, as the cost barrier fades, the superior characteristics of flash memory should prevail. It shouldn’t be long before datacenters bow to its performance, reliability, and sustainability benefits. And put the costs of acquisition, licensing, maintenance, support, scalability and eventual downtime into the pencil, you will see that all-flash solutions can generate significant savings.

In fairness: Older hybrid flash arrays used a combination of SAS and Serial Advanced Technology Attachment (SATA) controllers to connect to both hard drives and SSDs. Modern all-flash storage uses Non-Volatile Memory Express (NVMe) controller, which offers better. Not all solutions on the market are all-flash, so stay tuned.