Xilinx ISV Stand | Ruiwen Technology: Liberate the CPU with Open vSwitch
CPU is overloaded
Currently, the extensive business layer requirements have led to the rapid growth of data centers. The increasing number of virtual machines and containers, the ever-changing application requirements, and the increasing importance of network security cannot be solved by switches at all levels in the data center alone. Virtual machines once became a solution, but because this solution is based on software implementation, it occupies a considerable amount of CPU resources. The CPU that could have been used for business and applications is occupied, which in turn increases operating costs and energy consumption.
Smart NICs emerge
The contradiction between demand and software implementation has led to the emergence of smart network cards. Smart network cards are different from ordinary network cards. They are no longer just responsible for Layer 2 forwarding between the server and the outside world, but take on more functions to help the CPU handle network-related affairs, such as the virtual switch and virtual firewall mentioned above. They free up the CPU to handle more application needs.
As bandwidth increases, such as to 25G or even 100G, ordinary CPUs cannot meet the needs of network services, and they need to bind several CPU cores to perform network acceleration. Our solution can save CPU resources and offload network services performed by the CPU itself to the FPGA, thereby greatly improving CPU efficiency.
CTO of Ruiwen Technology
Yan Yan
xNIC reduces CPU load
Ruiwen Technology's xNIC series of smart network cards can reduce the burden on the CPU by simply replacing existing network cards, support programmable networks, and real-time network telemetry, thereby accelerating network applications.
At the 2019 XDF (Xilinx Developer Conference), Raymax CTO Ms. Yan Yan demonstrated the Open vSwitch offload solution based on the Alveo accelerator card. Compared with the CPU server, the system throughput increased by 1.4 times and the latency decreased by 7.6 times.
↓↓ Please watch the video for details↓↓