Amazon has released to beta a new web service. Charles Ying, after being under NDA, blogs some details. Amazon is running Erlang.
Greg Wilson, author of Practical Parallel Programming, picks up on the post and says this about pure functional languages:
I don’t believe that PFLs make non-trivial parallel programs easier to write. I don’t believe they make parallel programming harder, either, and the reason I don’t is that I haven’t seen any empirical studies of real programmers writing real programs that point either way.
via A Fresh Cup
{ 1 comment… read it below or add one }
I think that he’s got to come up with a more specific challenge than “make non-trivial parallel programs easier.” Obviously there are already cases where the concurrency model of Erlang or even pi-calculus greatly simplifies otherwise “non-trivial” applications, but clearly those aren’t the cases that Greg has in mind.
If he’s addressing people who peddle Erlang as the cure to all concurrency problems, he makes a good point (but then those people probably won’t even be listening, so it’s a wash).
I think the truth of the matter is that “concurrency” is too broad a domain to usefully talk about in such general terms. It’s true in sequential programs too that different notations help solve different problems — e.g.: reading assembly code is useful to immediately calculate the practical space and time costs of an algorithm, while writing in ML makes it easier to understand what a function is meant to compute altogether.