By J Metz and Bill Lynn
Now that we’ve talked through the history of NVMe and its building blocks, we’re ready to dive into form factors. Ultimately, this is the key point: Form factors dictate overall capability. In very real terms, form factors represent a physical consequence of the limitations that we’ve been talking about. Traditional drive form factors only support SAS and SATA.
The U.2 drive form factor supports a x4 PCIe interface, but fits into traditional 2.5” drive bay. M.2 modules support SATA or a x4 PCIe interface. EDSFF form factors can support a x4, x8, or x16 interface. PCIe adapter modules can support x1, x4, x8, and x16 PCIe interfaces.
Serviceability is another feature that needs to be considered when looking at form factors. U.2, EDSFF, and traditional drive form factors support external accessibility and hot plug. Form factors such as M.2 and PCIe adapter cards do not support external accessibility and hot plug.
Once you get past the host interface you now have to look at things like number of back-end ONFI channels, number of NAND chip placements, and module power. The larger the form factor, the more NAND chips you can support (hence greater capacity). The form factor also specifies the amount of power that can be supported. More power equates to a faster drive.
New Types of Form Factors
As the new technologies emerge into the marketplace, these trade-offs necessarily give way to new and interesting ways of developing storage solutions. NVMe has allowed a tremendous shift in the way that media is addressed so that there are new options and alternatives for creating capacity and/or performance in different packaging.
From “rulers” to new NVDIMMs (PDF), from new SSD interfaces to raw memory architectures, the ability to solve additional problems (space, power, cooling, density, reliability, etc.) warrants a greater degree of choice, as opposed to “making do” with what’s previously been available.
One of the new trends in form factors is adapting the new form factors in a carrier card. The most common instance of this is multiple M.2 SSDs on a PCIe low-profile add-in-card. One of the main advantages to this is to scale the amount of SSDs to allow for maximum bandwidth on the PCIe interface and scalable performance in bandwidth / TB.
The thing is, a PCIe x4 or x8 add-in card (“AIC”) in a x16 slot wastes many precious PCIe lanes. The very first PCIe SSDs were in the AIC form factor, but a carrier card with multiple SSDs allows hyperscale data centers to scale the performance by workload, mix and match for different server configurations, and have more smaller SSDs to isolate workloads for better quality of service and latency. These adapters will also allow innovation in new form factors like EDSFF while maintaining legacy compatibility.
We’ve explored a large number of criteria that influence decisions made for storage, but we have only scratched the surface. As new technologies emerge that make improvements – subtle or profound – we will be able to see how they apply to specific problem “sweet spots.”
If there is anything that we’ve discussed in this article that deserves greater attention, please feel free to let us know in the comments. There are a large number of topics that we only briefly covered, but are well-deserving of further exploration.
Ultimately, however, the emergence of new form factors for storage solutions is nothing to be concerned about. On the contrary, the ability to solve new and difficult problems in interesting ways is precisely one of NVMe’s greatest strengths, with more new and exciting technological answers to come.