Implementing code on demand using Erlang

Abstract

Erlang is an industry-strength programming language conceived for developing programs for highly concurrent and distributed settings. It follows the actor model of concurrency and offers constructs, tools and libraries that facilitate the construction of scalable, fault-tolerant distributed code. In particular, Erlang offers the gen_server behaviour, a framework for standard1smg and expediting the construction of distributed computation across nodes. This behaviour works well for certain instances of distributed configurations, such as traditional client-server arrangements, but fails to provide the necessary automation and support to facilitate other forms of distributed computing arrangements such as remote-evaluation and code-on-demand. Under pinning these problems is the fact that the gen_server behaviour assumes that each node in a distributed computation hosts the same code-set and resources. In this report I discuss an extension of the gen_server behaviour that provides additional functionality for the management and migration of un even code across distributed nodes. I package my solution as an extended behaviour while attempting to preserve the same interface of the existing gen_server, so as to facilitate adoption and maintain compatibility with existing legacy code. I show how this new behaviour can be used to simplify the implementation of distributed code architectures that are not handled well by the plain-vanilla gen_server behaviour, and discuss the performance impact of my solution.