Documentation
Rate This Document
Findability
Accuracy
Completeness
Readability
SubmitClose

Tuning Process

Tuning Environment

Table 1 Tuning environment

Environment

CPU

Memory

Storage

OS

Kernel Version

Application Deployment

Test environment 1

Kunpeng 920-4826 (128 cores)

4 × DDR4 32 GB 2933 MHz

HDD*2

openEuler 22.03 LTS SP4

5.10.0-153.12.0.92.oe2203sp2.aarch64

Docker container

Test environment 2

Kunpeng 920 7270Z (128 cores)

8 × DDR5 32 GB 4800 MHz

2.9T SSD*2

openEuler 22.03 LTS SP2

5.10.0-216.0.0.115.oe2203sp4.aarch64

Docker container

Test environment 3

Kunpeng 920-6426 (128 cores)

4 × DDR4 32 GB 2400 MHz + 10 × DDR4 32 GB 2933 MHz

1.1T HDD*2

openEuler 22.03 LTS

4.19.90-2003.4.0.0036.oe1.aarch64

Docker container

Test Method

  1. Prepare the test conditions.

    Deploy the application to a container and use a pressure test tool to evaluate its current performance. During the test, bind the application and pressure test tool to separate NUMA nodes to ensure resource independence and isolation. To avoid impacting the accuracy of test results, resources are not restricted for containers.

  2. Obtain the importance parameters and parameter values.

    On Kunpeng 920 servers, use the Kunpeng AutoTuner to perform comprehensive tuning and training based on test cases for various application scenarios. The tool identifies key performance-sensitive parameters and calculates their optimal and worst values, along with the corresponding performance under different test conditions. This data serves as a scientific basis for subsequent tuning.

  3. Verify the universality of importance parameters.

    To verify the universality of the identified sensitive parameters and their recommended values, the tool performs cross-verification in different Kunpeng 920 environments. The tool evaluates the optimal, worst, and baseline values, and compares data from multiple tests to verify that the selected parameters remain consistent and reliable across diverse hardware environments. Through this comprehensive verification process, the tool ensures that the recommended parameters and parameter values are both adaptable and reliable.