IoT applications are usually characterized by large-scale demand and the widespread use of mobile devices. Similarly, performing interaction among application and system components in a decoupled and elastic way, and enforcing Quality of Service (QoS) usually also become issues. Hence, paradigms such as pub/sub on top of cloud resources represent a suitable strategy for application development. However, management of QoS-aware resource allocation for pub/sub systems remains challenging, especially when system peers connect in an intermittent way. In this paper, we propose a new approach for resource allocation focusing on end-to-end performance in face of peers’ disconnections. We evaluate and demonstrate the benefits of our approach using simulations. QoS enforcement was achieved in almost all scenarios, and we have shown that our approach can help reasoning about efficient resource allocation.