✦ +
The purpose of big data performance testing
. Performance regression of big data components, perform performance comparison between new and old versions when the version is upgraded.
2. Obtain the performance baseline after the release of a new version/new production environment, establish a measurable reference standard, and provide a comparative reference for other test scenarios or tuning processes.
3. Test and comparison are carried out in many releases to provide reference data for PoC testing.
4. Supports POC testing and draws conclusions. At that time, different solutions can be selected based on business models and needs, or according to customer needs.
5. Perform performance testing on the customer's side to meet the performance standards required by the customer to meet the customer's needs.
✦ +
Shot of performance testing
- Online New version
- Online New environment/new host
- Online New area
- PoC Test
- Performance special test
✦ +
Step of performance testing
Definition of the target of the test
Determine the performance testing scenario, cluster size and specifications, data volume, data format, compression algorithm, etc.
For example:
- version iterative test needs to be aligned with the historical version cluster specifications and parameters to compare whether the version performance is deteriorated;
- Poc test needs to clarify the customer scenario;
- software publisher test needs to be consistent with the publisher cluster size.
Application host environment and test cycle
Build the running environment and monitor the performance indicators obtained by
, part of which includes: performance data, such as bandwidth, disk IO, CPU, memory and other indicators.
Test
During the test process, use nmon or other system monitoring tools to record system indicator changes to be discovered to discover system bottlenecks and facilitate subsequent tuning.
adjusts and optimizes performance results
performs iterative performance testing.
Product test report
✦ +
Hbasemtqt, in addition to the mainstream test tool Hibench, Yahoo's big data test suite.
✦ +
Big data performance tuning
There is a common problem of data tilt in the field of big data. It is necessary to refer to the official documents of the corresponding components and refer to the industry case introduction.
✦ +
Big data related test
Benchmark test
Single user single transaction test, the purpose is to obtain the system to process a single request without stress on the selected user.
Load test
Test the changes in system performance by gradually increasing the load of the system.
stability test
runs for 7*24 hours by loading the system with a certain amount of business pressure.
Functional test
Especially when selecting the OLAP engine, it is necessary to test its support for standard SQL, such as the department does not support update and delete operations, does not support with statements, does not support except and intersection operations, etc.
Performance requirements
CPU, memory, disk IO, and network load usage rate shall not exceed 80%, and the response time shall not exceed 3 seconds for 90% of the reading, writing, export and import shall not exceed 3 seconds, and less than 10% shall not exceed 5 seconds.
Test case
. Read, write, and export data import benchmark tests for Hadooph and spark respectively in different data volumes (100G, 500G, 1T).
2. Perform parallel and read-write mixed tests under different data volumes.
3. Perform 7*24 hours of data stability test under different data volumes.
Test observation indicator
- CPU usage
- memory usage
- IO
- network
- response time
- o
- response time
- ometric
