Hey, I picked that thread up through LtU, and in the thread you mentioned that you had a new blog there too, so that brought me here.

Interesting thread; I do dabble in Erlang. You mention that Erlang processes each have their own separate heap, and that there are no inter-process pointers. The trick reminds me of stories my professors have that back in the day, it was standard practice to simply reboot the computer when memory ran out instead of letting the collector do it's work.

My understanding is that Erlang processes, at least on a single node, share a common heap and pass messages via pointers. Wait... that's a ./configure option, IIRC, so I guess I answered my own question. Which option is most often used in actual deployments?

I suppose you could have a many-to-one function from processes to heaps, using an option to spawn to specify a heap, allowing the programmer to use both models in conjunction.

Anyway, here's a few of my journal entries you might enjoy: journey to unemployability, queues, and concurrency.

Take Care.

(Reply) (Thread)

Erlang historically used separate heaps for each process. More recently the HiPE group in Uppsala have made a shared-heap virtual machine (configure option) with different performance characteristics. I like to choose one behaviour and exploit it and that for me is separate heaps (partly because I'm stuck on OTP R9 at work). I also like to write code that annoys people in Uppsala (my favourite is applying {Mod,Fun} tuples) in petty retaliation for historical bickering with Kostis Sagonas (who is actually a very nice guy).

Nice blog and welcome to unemployability! I'm finding it's surprisingly lucrative. :-)

(Reply) (Parent) (Thread)