Evospace-js: Asynchronous Pool-based Execution of Heterogeneous Metaheuristics

This paper is part of a continuing effort in the field of EC to develop algorithms that follow an opportunistic approach to computing, allowing the exploitation of freely available services over the Internet by using free tiers of cloud services or volunteer computing resources; the EvoSpace model is able to tap from both kind of resources, using asynchronous evolutionary algorithms. We present its design, which follows an an event-driven architecture and asynchronous I/O model, and its implementation, with a server-side tier programmed in Node.js that uses Redis as an in-memory and high performance data store for the population. This population store is exposed to clients running population-based and nature-inspired metaheuristics through a REST API. Additional capabilities where implemented in this version to allow the logging of experiments where heterogeneous algorithms are executed in parallel. These logs can then be transformed to other formats. As a case study an hybrid global optimization algorithm has been implemented mixing two algorithms: a PSO algorithm from the EvoloPy library and a GA using the DEAP framework. The result was transformed to files compatible to the Comparing Continuous Optimizer platform in order to use their post-processing code. Clients in this case have been developed in the Python language, the language used to implement both libraries. The results from this case study suggest, first, that EvoSpace can be used as a paradigm- and language-agnostic platform for population-based optimization algorithms, and also that this software can yield performance improvements and a viable platform to execute and compare asynchronous pool-based metaheuristics.