Skip to content

Latest commit

 

History

History
109 lines (74 loc) · 4.76 KB

benchmark.md

File metadata and controls

109 lines (74 loc) · 4.76 KB

FreeBEE Benchmark and Functionality Test Harness

FreeBEE includes a module (freebee-bench) that provides a test harness that can be used for performance benchmarking and evaluator functionality testing.

The module includes a number of interfaces that allow the test harness to be configured or extended.

The module also includes useful companion classes, including:

  • A configurable, random boolean expression generator.
  • A reference boolean expression evaluator that is not optimized for performance but is easier to understand. (Note, this is a work in progress and currently has limited features.)

Usage

The benchmarking test harness currently consists of two Java main classes:

  • BenchmarkRunner: Stress tests the FreeBEE library and produces a runtime performance report.
  • ReferenceComparisonRunner: Compares the FreeBEE logic against a reference evaluator for correctness analysis.

BenchmarkRunner

The following steps run the benchmarking tool:

  1. Configure BenchmarkRunner. Currently, this is managed at the top of the main class as static variables. In the future, this will be replaced with external configuration properties. For now, modify these values in BenchmarkRunner.java:

    private static final int INPUT_COUNT = 5_000;
private static final int MAX_VALUES_PER_INPUT = 100;
private static final int EXPRESSION_COUNT = 1_000;
private static final int MAX_EXPRESSION_WIDTH = 1_000;
private static final int MAX_EXPRESSION_DEPTH = 3;
private static final long RANDOM_SEED = 14888790436548L;

    Here is the effect of these variables:

    Config Variable Description
    INPUT_COUNT The number of input records/requests to run against the evaluator.
    MAX_VALUES_PER_INPUT The maximum number of attribute values to generate per input record
    EXPRESSION_COUNT The number of boolean expressions to generate for the evaluator
    MAX_EXPRESSION_WIDTH The maximum number of predicate (leaf) nodes for the boolean expression trees that are generated
    MAX_EXPRESSION_DEPTH The maximum depth of the boolean expression trees that are generated
    RANDOM_SEED The random seed to use for all pseudorandom generators. Useful for reproducing results.

  2. Create an IDE Run Configuration for the BenchmarkRunner.java main class.

  3. Run the BenchmarkRunner Run Configuration. The output will go to the console. Here is an example:

     Evaluated 5000 requests against 1000 expressions
 
 Total time: 90414 ms
   Setup benchmark:        12440 ms
   Build evaluator index:  2282 ms
   Evaluate requests:      75692 ms
 
 Average and Max Metrics
   Average evaluation time:  15.138400 ms (wall clock)
   Average evaluation time:  15.133384 ms (instrument)
   Standard deviation:       12.868864 ms
   50th percentile:          13.676861 ms
   75th percentile:          19.150101 ms
   95th percentile:          28.939144 ms
   97th percentile:          33.324185 ms
   98th percentile:          38.730450 ms
   99th percentile:          51.663963 ms
   Max evaluation time:      572.646660 ms
 
 Request Input Stats
   Matched:      5000 requests
   Not matched:  0 requests

    Additional detailed output will be logged to stderr.

ReferenceComparisonRunner

The following steps run the reference comparision tool:

  1. Configure ReferenceComparisonRunner. Currently, this is managed at the top of the main class as static variables. In the future, this will be replaced with external configuration properties. For now, modify these values in ReferenceComparisonRunner.java:

    private static final int INPUT_COUNT = 100;
private static final int MAX_VALUES_PER_INPUT = 100;
private static final int EXPRESSION_COUNT = 100;
private static final int MAX_EXPRESSION_WIDTH = 100;
private static final int MAX_EXPRESSION_DEPTH = 4;
private static final long RANDOM_SEED = 14888790436548L;

    For the details regarding the meaning of these variables, see the above section for BenchmarkRunner.java

  2. Create an IDE Run Configuration for the ReferenceComparisonRunner.java main class.

  3. Run the BenchmarkRunner Run Configuration. The output will go to the console. Here is an example:

     Evaluated 100 requests against 100 expressions
 
 Total time: 14512 ms
   Setup benchmark:        1708 ms
   Build evaluator index:  569 ms
   Evaluate requests:      12235 ms
 
 Expression evaluation results
   Total failed expressions:  0 (0.0)%

    Additional detailed output will be logged to stderr.