Erlangen: asynchronous, distributed message passing for Clozure CL (lightning talk for ELS 2017)

• asynchronous message passing
• reliable (supervision trees)
• inherently distributed
• inspired by Erlang, yet very different

• uses native OS threads (can’t have 10,000)
• does not enforce strict isolation (within the boundary of a node)
• Common Lisp is synchronous (blocking I/O)

…there are also many reasons to prefer CL/Erlangen over Erlang

• for instance, Erlangen’s message queues are bounded and never leak memory

• uses native OS threads

even if processes are heavy, supervision trees are worth their weight in gold

• does not enforce strict isolation

just don’t modify objects you send (send = free)

• Common Lisp is synchronous (blocking I/O)

tempted to look at asynchronous I/O (IOLib): complete dead end, no tasteful way to write code

blocking is fine as long as you always supply a timeout

• 5us for message delivery, 10x more than erts 7.3 (that’s not even bad, actually)
• up to 2 million messages per second
• you can have 1,000 processes easily

…no serious optimization work done so far!

• portability layers are boring
• portability layers are boring
• portability layers are boring
• I won’t write one or maintain one
• Erlangen already includes modifications to CCL
• …maybe I want to swap out CCL’s threading, GC, I/O, … implementations?
• erlangen-kernel is distributed as an executable that includes Quicklisp

• github.com/eugeneia/erlangen AGPL-3.0
• branch, fork all you like
• nothing is set in stone yet, wild ideas welcome
• mail me at max@mr.gy