gonutz / performance_aware_programming

See https://www.computerenhance.com

Performance-Aware Programming

Course by Casey Muratori at

https://www.computerenhance.com

Haversine Distance Problem

First generate the test data file by running gen_test_data.go.

cd haversine
go run gen_test_data.go

This will create data_10000000_flex.json. This is necessary for the other code to run.

The original Python code (listing 21), presented in the course is in haversine/haversine_python.

I rewrote the Python code in Go, not changing it, only adopting it to the language. See haversine/haversine_go.

Finally, I tried my best to create the fastest Go version possible. See haversine/haversine_go_fast

Python:
	Result: 10007.8833106713
	Input = 17.622838020324707 seconds
	Math = 19.40399718284607 seconds
	Total = 37.026835203170776 seconds
	Throughput = 270074.39186008676 haversines/second

Go:
	Result: 10007.8833106713
	Input = 17.8604208s
	Math = 1.1329996s
	Total = 18.9934204s
	Throughput = 526498.1129991731 haversines/second

Optimized Go:
	Result: 10007.88331067118
	Threads = 8
	Count = 10000000
	Total = 377.4745ms
	Throughput = 26491856 haversines/second

The optimized Go version is about 50 times faster than the original Go version and about 100 times faster than the original Python version.

Note that due to a different order of summation, rounding errors propagate differently for the optimized version, thus we get a slightly different result.