Specifications

vKAE supports hardware acceleration on VMs allocated by servers powered by Kunpeng 920 processors. vKAE supports multiple VM specifications, such as 4C8G/8C16G/16C32G/32C64G, and is also applicable in container scenarios.

In the VM specifications, "4C8G" indicates that 4 vCPUs and 8 GB memory are allocated to the VM. This rule also applies to other specifications.

• vKAE performs hardware acceleration for the RSA algorithm using the VF passthrough to VMs. The openssl speed command is used to test the performance. The test result shows that the performance is tripled after vKAE is used. The improved performance is as good as that improved by KAE on a physical server. Compared with using KAE on a physical server, using vKAE on a VM has no performance loss.
• In the OpenSSL application environment, the openssl speed command is used to test the performance on a VM that is configured with vKAE. It is found that the performance trend is directly associated with the number of vCPU cores. In the synchronous or asynchronous operation mode, the performance upper limit of RSA-sign highly depends on the number of vCPU cores allocated to the VM.

  In a VM with 64 vCPU cores, the maximum throughput of the RSA-sign metric reaches about 54,000 signs per second. When the number of vCPU cores is doubled to 128, the maximum throughput of RSA-sign also doubles, reaching about 108,000 signs per second. This proves the association between performance improvement and CPU resource expansion.

• vKAE supports the asynchronous mode of Nginx 1.21.5 (and later versions), which aims to provide hardware acceleration for the RSA algorithm during HTTPS handshake process, thereby improving the performance.
• Tests are performed using the httpress tool on an 8C16G VM in the Nginx scenario. It is observed that the CPU performance does not reach the bottleneck with the enablement of vKAE, and the asynchronous mode supported by vKAE demonstrates better performance than that of the synchronous mode: The request per second (RPS) in the asynchronous mode is 40% higher than that in the synchronous mode, and the average response time is 30% shorter.