By Marc Cram, Server Technology
Persistent memory can benefit the data center in that it is nonvolatile memory – the memory contents are not lost even when power is removed. Its technologies, such as Ferroelectric RAM (FRAM), Magneto-resistive RAM (MRAM), and others, deliver faster access speeds than rotating media with lower power consumption on a per-bit basis. These also tend to be smaller in size and generate less heat, meaning they can be applied in more places physically and require less cooling to keep them running. Persistent memory also delivers a very reliable ability to turn a system completely off, taking power consumption to zero, and then restoring that machine to full operation.
Persistent memory is perfect for large datasets that require faster access in order to provide meaningful fiscal outcomes. This includes, for example, high-frequency trading algorithms, editing and rendering for animated films, and scaling up servers where multiple virtual machines (or containers) will be running concurrently and need to be able to switch between them rapidly.
In addition to the FRAM (ferroelectric memory) and MRAM (magnetic memory) mentioned above, there are many other persistent memory technologies, such as NVDIMM-N – which combines DRAM and DIMM on a module, addressed like DRAM, backed up to NAND periodically. There is also NVDIMM-F – a “flash only” solution, basically flash sitting on the memory bus, slower than DRAM; PCM – Phase Change Memory; Crossbar technology – Intel Optane/ 3D XPoint; and RRAM/ReRAM. The chart below shows where persistent memory most often sits within the hierarchy of storage within a server system.
The persistent memory solution is likely to become a widely available technology when the cost per bit comes down, and this usually happens as they climb the curve of gate density per square millimeter on a wafer, combined with the development of high-value use cases.
You can prepare for persistent memory in your data center by talking with your systems vendors to find out their roadmaps or exploring your workloads to determine applications that would benefit from the speed/cost trade-off for persistent memory. You can also review network capacity to ensure available bandwidth is sufficient to accommodate systems incorporating persistent memory, migrate your workloads, and optimize your operating environment – i.e. raising the data center temperature.
About the author:
Marc Cram is Director of New Market Development for Server Technology (@ServerTechInc), a brand of Legrand (@Legrand). Marc is driven by a passion to deliver a positive power experience for the data center owner / operator. Marc earned a BSEE from Rice University, and has over 30 years of experience in the field of electronics.