Mid Price Benchmarking

Table of Contents

Overview
Benchmark Methods
Results
Usage
How it Works
License
Conclusion

Overview

This project benchmarks different methods for calculating the mid price of two values using different numeric types: double, BigDecimal, Decimal64 and Decimal6f. It compares the performance of these methods with different rounding strategies, including rounding to six decimal places.

The benchmarks are implemented using the Java Microbenchmarking Harness (JMH). It tests the speed and accuracy of mid-price calculations on a fixed-size data set.

Benchmark Methods

The following mid-price calculation methods are compared:

Double Precision (with rounding): Uses primitive double types to calculate the mid-price and rounds to 6 decimal places.
Double Precision (without rounding): Uses primitive double types to calculate the mid-price without any rounding.
BigDecimal (Divide by 2): Uses the BigDecimal class to calculate the mid-price by dividing the sum of two values by 2 and rounding to 6 decimal places.
BigDecimal (Multiply by 0.5): Uses the BigDecimal class to calculate the mid-price by multiplying the sum of two values by 0.5 and rounding to 6 decimal places.
BigDecimal (Multiply by 0.5 without rounding): Uses the BigDecimal class to calculate the mid-price by multiplying the sum of two values by 0.5 without rounding.
Decimal64 (with rounding): Uses Decimal64 (from the deltix library) to calculate the mid-price and rounds to 6 decimal places.
Decimal6f: Uses Decimal6f (from the tools4j library) to calculate the mid-price with rounding. decimal4j on GitHub

Results

The benchmark was run on a system with the following configuration:

Processor: AMD Ryzen 5950X
Operating System: Linux
Java Version: Azul JDK v23

Benchmark Results (measured in operations per second):

Benchmark	Score (ops/s)	Error (ops/s)
`MyBenchmark.bigDecimalMidPriceDivide`	83,467	± 529
`MyBenchmark.bigDecimalMidPriceMultiply`	90,053	± 785
`MyBenchmark.bigDecimalMidPriceMultiplyWORounding`	114,612	± 963
`MyBenchmark.deltixDecimal64MidPrice`	63,605	± 434
`MyBenchmark.decimal6fMidPrice`	181,474	± 2,444
`MyBenchmark.doubleMidPrice`	855,706	± 3,239
`MyBenchmark.doubleMidPriceWORounding`	9,751,458	± 782,845

Usage

To run the benchmark, you can execute the following command:

mvn clean install
java -jar target/benchmarks.jar

This will compile and run the benchmark suite, showing results for each of the methods being tested.

How it Works

The benchmarking class, MyBenchmark, initializes three sets of data representing the same values using double, BigDecimal, and Decimal64. The calculations are done using different mid-price calculation techniques.

The results for each type are stored in separate arrays (ap, bp for double, ap2, bp2 for BigDecimal, and ap3, bp3 for Decimal64). After calculating the mid-prices, the values are compared across the different data types to ensure consistency.

License

This project is licensed under the Apache 2.0 License.

Conclusion

This benchmarking suite provides valuable insights into the performance and accuracy of different numeric types and calculation methods in Java. The results indicate that double and BigDecimal are typically faster than Decimal64 for simple mid-price calculations, but BigDecimal may offer more precision and flexibility for financial and high-accuracy computations.

For more details on JMH and performance benchmarking, refer to the official JMH documentation