Language Benchmarking
🏃 TLDR: Based on these benchmarks and my code to aggregate the CPU seconds column, Go and Java perform the same, C# performs twice as fast as both of them, Rust outperforms C# by a small margin. Python is 10x slower than Go or Java.
My favorite bench marking is the Computer Language Benchmarks Game created by Debian developers. They test physics and astronomy simulations (N-body), various matrix algorithms, binary trees, regex, and more. Though you can do your best to interpret their mean results. I have written some awk and sed to provide another simpler metric comparing one language to another.
Take this this data comparing Go and Python which I've copied directly from the website.
fannkuch-redux
source secs mem gz cpu secs
Go #3
8.25 10,936 969 32.92
Go
11.83 11,128 900 47.26
Go #2
11.89 12,708 896 47.47
Python 3 #6
913.87 11,080 385 913.83
Python 3 #4
285.20 14,264 950 1,123.47
n-body
source secs mem gz cpu secs
Go #3
6.36 11,244 1200 6.37
Go
6.55 11,244 1310 6.56
Go #2
6.94 11,244 1215 6.95
Python 3
383.12 11,096 1196 383.11
Python 3 #2
402.59 11,096 1242 402.57
Processing the Data
I only care about the secs to measure CPU time. So that means I can remove the names of the algorithms as well as the headers that include source, secs, etc. A single mention of source will remove the each entire record. Though there are none above, there are some tests that fail, we'll remove those and the line above them that contain Python or Go.
/source\|binary\|reverse\|nucleotide\|fasta\|regex\|pidigits\|mandelbrot\|spectral\|n-body\|fannkuch/d
N; /Timed\|Bad\|Failed/{d;}; P; D
Now before I run my awk script to average out the first numbered field, I want to make sure Python and Go are on the same record as the numbers. In vim, J will remove a newline, basically moving current selected line up one. Let's run that on every line. Also the test numbers #3 I want to remove as well as the 3 after Python. That will make it easier for awk to read just the data we want.
After all that our data should look much cleaner.
Go 33.37 247,824 525 60.85
Go 34.62 236,728 482 61.41
Python 104.73 271,980 338 104.72
Python 35.33 274,884 660 125.79
Now using awk I simply take the second field and average them for Go and Python separately.
awk '/Python/{python_sum+=$2; python_count++} /Go/{go_sum+=$2; go_count++} END{print "Python Average: " (python_sum/python_count); print "Go Average: " (go_sum/go_count)}' go-python.txt
Final Results
Here is the result. I've also included some other languages as well. First I compare Go to three other languages, then C# against Java and Go in the last two. Not surprising to see Go 10x faster than Python, though I was shocked to see C# performed 2-3x faster than Go and Java.