We measured the all-to-all bandwidth in two different scenarios: in the first one, which we call plain, threads directly send back-to-back, coarse-grained messages to random target nodes, while in the second one, which we call aggregated, fine-grained messages are used, and they are sent using our a2a API.