Co-ops: concurrent algorithmic skeletons for Erlang

Abstract

Erlang offers a programmer 3-4 orders of magnitude more processes than conventional languages. This difference in approach to concurrency leads to architectures and attitudes embracing processes as key elements of a software system, providing fault isolation, distributed algorithms, and code modularity. Coming hardware improvements promise 3-4 orders of magnitude more CPUs than conventional hardware. How will this additional power be used by Erlang programmers and how might it impact Erlang system architecture? This poster introduces a new library of cooperating processes or "co-ops" which implement an algorithmic skeleton as a directed acyclic graph (DAG) spanning a large number of processes, trading program code for dataflow scaffolding to gain a more principled architecture and explicitly defined concurrent data pathways.