Doris BE Storage Layer Benchmark Tool

usage

It can be used to test the performance of some parts of the BE storage layer (for example, segment, page). According to the input data, the designated object is constructed, and the google benchmark is used for performance testing.

Compilation

  1. To ensure that the environment has been able to successfully compile the Doris ontology, you can refer to [Installation and deployment] (https://doris.apache.org/master/en/installing/compilation.html)。

  2. Executerun-be-ut.sh

  3. The compiled executable file is located in ./be/ut_build_ASAN/test/tools/benchmark_tool

operator

Use randomly generated data set for segment read test

The data set will be used to write a segment file first, and then the time-consuming scan of the entire segment will be counted.

./benchmark_tool —operation=SegmentScan —column_type=int,varchar —rows_number=10000 —iterations=0

The column_type here can set the schema, the column type of the segment layer currently supports int, char, varchar, string, the length of the char type is 8, and both varchar and string types have length restrictions Is the maximum value. The default value is int,varchar.

The data set is generated according to the following rules.

int: Random in [1,1000000].

The data character set of string type is uppercase and lowercase English letters, and the length varies according to the type.

char: Length random in [1,8]。 varchar: Length random in [1,128]。 string: Length random in [1,100000]。

rows_number indicates the number of rows of data, the default value is 10000.

iterations indicates the number of iterations, the benchmark will repeat the test, and then calculate the average time. If iterations is 0, it means that the number of iterations is automatically selected by the benchmark. The default value is 10.

Use randomly generated data set for segment write test

Perform time-consuming statistics on the process of adding data sets to segments and writing them to disk.

./benchmark_tool —operation=SegmentWrite

Use the data set imported from the file for segment read test

./benchmark_tool —operation=SegmentScanByFile —input_file=./sample.dat

The input_file here is the imported data set file. The first row of the data set file defines the schema, and each row corresponds to a row of data, and each data is separated by ,.

Example:

  1. int,char,varchar
  2. 123,hello,world
  3. 321,good,bye

The type support is also int, char, varchar, string. Note that the data length of the char type cannot exceed 8.

Use the data set imported from the file for segment write test

./benchmark_tool —operation=SegmentWriteByFile —input_file=./sample.dat

Use randomly generated data set for page dictionary encoding test

./benchmark_tool —operation=BinaryDictPageEncode —rows_number=10000 —iterations=0

Randomly generate varchar with a length between [1,8], and perform time-consuming statistics on encoding.

Use randomly generated data set for page dictionary decoding test

./benchmark_tool —operation=BinaryDictPageDecode

Randomly generate varchar with a length between [1,8] and encode, and perform time-consuming statistics on decoding.

Custom test

Here, users are supported to use their own functions for performance testing, which can be implemented in /be/test/tools/benchmark_tool.cpp.

For example:

  1. void custom_run_plus() {
  2.     int p = 100000;
  3.     int q = 0;
  4.     while (p--) {
  5.         q++;
  6.         if (UNLIKELY(q == 1024)) q = 0;
  7.     }
  8. }
  9. void custom_run_mod() {
  10.     int p = 100000;
  11.     int q = 0;
  12.     while (p--) {
  13.         q++;
  14.         if (q %= 1024) q = 0;
  15.     }
  16. }

You can join the test by registering CustomBenchmark.

  1. benchmarks.emplace_back(
  2.                     new doris::CustomBenchmark("custom_run_plus", 0,
  3.                         custom_init, custom_run_plus));
  4. benchmarks.emplace_back(
  5.                     new doris::CustomBenchmark("custom_run_mod", 0,
  6.                         custom_init, custom_run_mod));

The custom_init here is the initialization step of each round of testing (not counted as time-consuming). If the user has an object that needs to be initialized, it can be implemented by a derived class of CustomBenchmark. After running, the results are as follows:

  1. 2021-08-30T10:29:35+08:00
  2. Running ./benchmark_tool
  3. Run on (96 X 3100.75 MHz CPU s)
  4. CPU Caches:
  5.   L1 Data 32 KiB (x48)
  6.   L1 Instruction 32 KiB (x48)
  7.   L2 Unified 1024 KiB (x48)
  8.   L3 Unified 33792 KiB (x2)
  9. Load Average: 0.55, 0.53, 0.39
  10. ----------------------------------------------------------
  11. Benchmark                Time             CPU   Iterations
  12. ----------------------------------------------------------
  13. custom_run_plus      0.812 ms        0.812 ms          861
  14. custom_run_mod        1.30 ms         1.30 ms          539