The first application of our technology will be a Top-of-Rack (ToR) system which functions as the high-speed disaggregated memory bank for all of the servers deployed in that rack. Our product connects to these servers using current and cutting-edge high-speed interconnection technologies like PCIe 5.0, which are typically used for communication between low-latency, high-bandwidth server components connected to the same motherboard.
Our first three products will use a range of interconnection and memory access technologies to accommodate various budgets and performance needs, including PCIe 5.0. We use custom FPGA acceleration to make our performance as close as possible to what each server expects from its local memory, which also gives us the flexibility to target both existing and new interconnection and memory access technologies that emerge over time.
Our technology is unique in that due to various physical and logical layer challenges, PCIe 5.0 and similar technologies were not designed to interconnect devices which were not connected to the same motherboard. We are effectively extending this low-latency, high-bandwidth interconnect and interfacing it directly with the memory access interfaces of the banks of DRAM inside our product at the top of the rack.
This allows each server within the rack to access memory at speeds comparable to their own local memory, which in turn allows for each server to have less dedicated local memory, through memory disaggregation.
These real images of our product show the standard 19″ 2 RU form factor which will be familiar to anyone who has deployed equipment in a data center. The backplane of the device hosts the high-speed connectivity to the rest of the rack, and each memory bank can be swapped out at the front. From a hardware perspective, dual hot-swappable power supplies are available, as are interfaces for network connectivity and local admin access.
Our product is supported by a robust memory manager which allows the administrator to monitor, control and optimize the memory allocation given to any particular client server in the rack in real-time. We believe that visibility for the administrator to the real-time memory usage behavior of specific workloads will allow for the most effective use of our disaggregated memory technology at data center scale, and so it is a focus of our development.
Benchmarks and Comparison
We believe that we are the farthest along in solving the challenge of one memory for all. As we refine our design and product capabilities, we will share performance figures with the industry that validate our unique approach to solving the problem, and the value of solving the problem itself. Stay tuned.
Why Disaggregated Memory?
One memory for all. A single disaggregated bank of high-speed memory for the whole rack.
Our technology enables In-Memory Computing (IMC) at data center scale.
Optimize cost, accelerate current workloads, and enable new ones with us.