Hybrid Implementation Techniques in Erlang BEAM

Erlang is a concurrent programming language designed for prototyping and implementing reliable real-time systems. Erlang BEAM [1] is an efficient and portable sequential implementation of Erlang, where Erlang programs are compiled into the C language.

Compiling into C gives very good portability, and low-level, hardware specific optimization, and allows for stand-alone applications. At the same time it introduces two-steps compiling, and large object code sizes. In contrast to compiling into C, threaded code emulation provides small code sizes and short turn-around times.

In this paper we describe a hybrid system based on Erlang BEAM mixing threaded code emulation with compiling into C. When constructing the new system we made the following decisions: mixing C code with threaded code should not degrade the C code performance, and there would be no restrictions on mixing the two execution models.

Performance of the new hybrid system was compared with: KLIC, a system compiling KL1 source code into C; JAM, an Erlang byte code interpreter; and VEE, an Erlang threaded code emulator.

KLIC generally performs better than Erlang BEAM C code, mainly due to differences between the two languages. Both Erlang BEAM C code and threaded code perform better than JAM and VEE. Erlang BEAM produces the smallest amount of compiled threaded code. Compiling into Erlang BEAM threaded code is about 15-22 times quicker than compiling into C (both for KLIC and BEAM).

The presented evaluation results show that mixing compiling into C with threaded code emulation gives full control over the required performance and code size, and provides short turn-around times allowing for incremental programming.