Simplifying Hands-On Teaching of Distributed Algorithms with SPLAY

Teaching distributed algorithms using a learning-by-doing approach is usually associated with a slow and cumbersome learning process for students. In order to test and evaluate even simple protocols, students need to learn how to set up and operate a testbed, and to write scripts for deploying, running their program and finally retrieving some logs for parsing and observation of the results. Moreover, the amount of code required in languages such as C# or Java for implementing even a simple protocol is often one order of magnitude more than the length of the protocol pseudo-code description as discussed in class or in research papers. Nevertheless, teaching distributed algorithms and protocols using real deployed code on real conditions is highly desirable and can not always be replaced by the use of simulators. We present in this paper our experience of using SPLAY, a distributed systems evaluation framework that greatly simplifies the work of both instructors and students for hands-on learning of distributed systems. SPLAY simplifies the writing, deployment and observation of distributed algorithms and the management of test environments, narrowing the complexity gap between pseudo-code descriptions and executable implementations. In addition, SPLAY's features and the focus kept on the algorithms and their evaluation, allow students to evaluate their protocols in a variety of conditions, by controlling the experiments and their running conditions, or by allowing running them on multiple testbeds at no additional costs and with minimal administration complexity.