Greenlets provide concurrency but not parallelism. Concurrency is when code can run independently of other code. Parallelism is the execution of concurrent code simultaneously. Parallelism is particularly useful when there's a lot of work to be done in userspace, and that's typically CPU-heavy stuff. Concurrency is useful for breaking apart problems, enabling different parts to be scheduled and managed more easily in parallel.
Greenlets really shine in network programming where interactions with one socket can occur independently of interactions with other sockets. This is a classic example of concurrency. Because each greenlet runs in its own context, you can continue to use synchronous APIs without threading. This is good because threads are very expensive in terms of virtual memory and kernel overhead, so the concurrency you can achieve with threads is significantly less. Additionally, threading in Python is more expensive and more limited than usual due to the GIL. Alternatives to concurrency are usually projects like Twisted, libevent, libuv, node.js etc, where all your code shares the same execution context, and register event handlers.
It's an excellent idea to use greenlets (with appropriate networking support such as through gevent) for writing a proxy, as your handling of requests are able to execute independently and should be written as such.
Greenlets provide concurrency for the reasons I gave earlier. Concurrency is not parallelism. By concealing event registration and performing scheduling for you on calls that would normally block the current thread, projects like gevent expose this concurrency without requiring change to an asynchronous API, and at significantly less cost to your system.
与恶龙缠斗过久,自身亦成为恶龙；凝视深渊过久,深渊将回以凝视…
