
Vendors and consumers, alike, have struggled with the balance of the amount of available RAM vs. disk. HPE has teased the long-awaited Memristor, and Western Digital has shown off ReRAM drives as potential solutions, but Intel’s Optane is the only category of products that promise to blur the lines between RAM and disk.
One can’t overstate the potential impact of these technologies on the data center. Even as a young programmer with a rudimentary understanding of memory and storage buses, I understood the impact.
Historically challenging
As a young PC enthusiast, I quickly learned the value of RAM vs. disk-based storage. As a hobby, I wrote a slide animation application in BASIC. The application was incredibly simple. I created an editor that allowed the user to save each frame individually. When the user hit play, the app would load each frame from storage, overwriting the previous frame.
Initially, the animation was choppy. The issue was the latency of reading each image from disk. As a workaround, I discovered the ability to create a RAM disk, thus creating a poor man’s cache system. I could pre-stage the animations to RAM, resulting in a smooth animation. The animation was so fast that I had to place a delay between each render.
Over the years, I’ve had the inverse problem as well. Some applications required more memory than available in my system. DRAM is cost prohibitive, and disk storage wasn’t practical to address directly. For years, the industry has teased the concept of a randomly addressable storage subsystem fast enough to use as short term system memory. Just as in my animation use case, the system didn’t need to provide memory at DRAM speeds, but just fast and cheap enough for practical purposes.
3D XPoint – A new type of memory
Intel recently announced the availability of the long-awaited Optane drive. Optane leverages 3D XPoint, jointly developed by Intel and Micron to solve basic computer I/O memory constraints. 3D XPoint, as a permanent storage medium, was to be magnitudes faster than current SSD and slower, but much cheaper, than DRAM.
As noted, the first drives to use 3D XPoint memory are the Intel Optane brand of drives. While not magnitudes faster, the drives are some of the fastest SSDs on the market. However, the ability to use the drive as addressable RAM is more intriguing for data center managers.
The Enterprise Data Center
My simple example app serves as an illustration of the challenges faced in the enterprise data center. The number of memory slots and cost limit the amount of DRAM a server holds. An Intel Xeon server filled with Optane could potentially hold as much as 24TB of 3D XPoint-based RAM. An obvious use case is in-memory databases. The ability to have several TB of data in-memory significantly reduces query times and results in faster business answers.
SEE: What would a memory-centric system look like? (ZDNet)
Generally speaking, virtualized and hyperconverged infrastructures should benefit from 3D XPoint memory as well. Very few virtualized environments experience CPU resource constraints. Memory and disk I/O are the typical constraints. With tweaks to the hypervisor, virtualized infrastructures could take advantage of different levels of RAM. For example, OS instances could run in DRAM. The virtual machine tools may instruct the hypervisor to make 3D XPoint available for application memory vs. DRAM.
I need to temper your excitement. Optane’s 375GB drive is in limited supply until the third quarter. The limited availability means customers looking to load up on Optane for some of these advanced use cases must wait a bit longer, until capacity is ramped up and the ecosystem figures out how to optimize applicaitons and systems.