Speed & Optimization - duxiaoyao/pdp GitHub Wiki

Python code is VERY SLOW and VERY POOR MULTI-CORE CONCURRENCY

• dynamically-typed
• GIL (Global Interpreter Lock) per process

Python+uvloop:

• faster than Golang on a single core but does not scale on two cores (单核networking很强，不输Go，但多核不行)
• Its performance drastically decreases when running python code to process the request

Python with Cython scales on multiple cores and runs faster than Go

Cython + wrapper of C libraries
可以通过Cython等C化缓解: statically compiled to C, release GIL for multicore concurrancy

Use Rust, Go, or C/C++: low latency, high concurrency, multi-core execution within single user-space shared memory

Back to the Metal:

• Rust to replace C/C++
• Go to replace Java

对Web和微服务（IO密集、无状态）来说，通过利用asyncio和起多进程，Python代码运行性能的问题不是问题。

References:

1. Multi-core Python HTTP faster than Go