Towards an Adaptive and Federated Testbed for AI Research in Africa

In a low-resource setting, unique challenges such as limited infrastructure, high network costs, and unpredictable network connectivity can hinder the setup and operation of research testbeds. Often, researchers rely on local and limited institutional resources or distant, less accessible platforms in other continents. However, a federated research testbed can provide an expanded resource pool with unified administrative features, promoting collaboration and contextual research in areas such as Artificial Intelligence (AI) and protocol design that align with local needs. In this paper, we propose a federated compute systems testbed spanning Research and Education Networks in Africa that is adaptive to the challenges unique to computing in Africa. Five key design considerations for such a testbed are presented: Adaptive, Ubuntu, Replicability, Ease-of-use and Secure and Policy-driven. The main ideas put forth in this paper are that a federated and topology-aware network and systems testbed anchored in NRENs can lower the costs of AI research and serve as a practical workaround for the prevalent skill and infrastructural gaps in Africa.