NVMe over Fabrics P2P Reference Architecture |
|
Overview
Reference Architecture for NVMe over Fabrics ApplicationsIn conjunction with its Accelerate Ecosystem partners, Microsemi has developed a unique reference architecture for NVM Express over Fabrics (NVMe-oF) applications. Standard NVMe-oF forces all IOs through CPU DRAM, putting unnecessary strain on the control plane and CPU. NVMe-oF with PCIe peer-to-peer (P2P) allows data IOs to bypass CPU DRAM and flow directly between NIC and NVMe endpoint which is achieved by:
As shown in the block diagram, the yellow dotted line represents data path without P2P from a NVMe JBOF all the way to the client CPU via server CPU’s DRAM. This burdens the server CPU especially in data-intensive applications. On the other hand, the green dotted line represents the same data path with P2P enabled, bypassing server CPU’s DRAM completely. This provides the benefit of using a smaller server CPU or using the freed-up server resources to run other applications. The plot to the left shows the CPU DRAM usage (in green) significantly goes down in a P2P scenario compared to a non-P2P, keeping the data copying speed (in orange) remains constant in both the scenarios.
FBOF stands for NVMe over Fabrics attached just a Bunch of Flash. FBOFs can be created by combining JBOF, Fabric, and P2P as follows:
Key Microsemi Products
Key Applications
More information on NVMe-oF P2P reference architecture can be accessed from below resources To learn more about this NVMe-oF P2P reference architecture, contact your local Microsemi sales office. |
This part can be found in the following product categories: