I don't think Stackless is nearly as well-known as it should be. It has a solid value proposition and is blazing fast in my experience as well.

Stackless is missing some important parts. I've spent a great deal of time working with it over the past few months, and there's no good SSL socket support. The built-in socket module will block the entire process (all tasklets), and while drop-in replacements exist, they either don't work with SSL, are buggy, or both.

Stackless also lacks Erlang's transparent support for multi-process and multi-machine execution, though I talked to a guy who's working on adding that for his Master's degree.

I'll be the first to admit that http socket issues exist and drop-in replacements are indeed often buggy. That being said, this wouldn't be a huge issue to overcome were Stackless given more TLC. Regardless, it has many other valuable uses.

For most things language performance is just a constant - the really hard issue is making something distributed and it's really here that Erlang shines. Being a Python programmer I know that creating a distributed application in Python is far from simple - while it's almost a trivial task in Erlang.

I was wondering about this. Is it really comparing like with like? I do not know Erlang, but I thought it had a lot of overhead from being designed for distributed apps etc.

A better comparison would be with Multiprocessing, or Eventlet/Spawning.

It's certainly important to know where Erlang doesn't perform so well. Project Euler ( http://projecteuler.net/ ) is also good for exploring the strengths of a given lanaguage. In terms of building robust and scalable distributed systems it cannot be touched. This stems for it's motivating use case of telephony. I'm not sure about Python but I know Haskell has been used largely in academic circles and hasn't had the same amount of attention paid to robustness. Joe Armstrong wrote a very good paper on the history of Erlang's development ( http://lambda-the-ultimate.org/node/2811 ).

Sounds like the thread-ring benchmark (http://shootout.alioth.debian.org/u32/benchmark.php?test=thr...) at the shootout. It benchmarks an Erlang program that isn't the first the author ever wrote, so its performance is probably more representative.

Yep - and both derive from this 10 year old measurement of context switching:

"Performance Measurements of Threads in Java and Processes in Erlang" http://www.sics.se/~joe/ericsson/du98024.html

Article is from Summer 2007.

as was already pointed out, this isn't a very accurate benchmark; it wasn't tested on a multi-core/processor machine, nor more importantly was it tested on a cluster of machines.

not surprising that python is competitive...afaik haskell also crushes erlang in lightweight threading benchmarks (supposedly the core strength of erlang).

Have a link for this? But lightweight threading isn't all that Erlang brings to the table. It comes with an entire library designed to create extremely fault tolerant, live-upgradable, extremely concurrent systems.

Getting lightweight threading of the same speed or faster is good, but it's less than 1/3 of the equation.

Fault-tolerance (uptime) is Erlang's core strength. Concurrency is just a pleasant and currently trendy side benefit.

> afaik haskell also crushes erlang in lightweight threading

For some definition of "crushes"

http://shootout.alioth.debian.org/u32/benchmark.php?test=thr...

But with quad-core Erlang processes float between cores slowing things down for sequential programs

http://shootout.alioth.debian.org/u32q/benchmark.php?test=th...

Nice article. Anybody want to try this with the new CPython multiprocess module?