Experiences on implementing a QoS-aware architecture on a DHT based broker overlay

Abstract

The new scale of the distributed systems due to the evolution of the Internet and new emergent technologies has given rise to applications based on the dissemination of high volumes of information. Future software systems must be responsive to events and must be able to adapt the software to enhance business processes. In this context, the publish/subscribe paradigm has taken special relevance. Event Broker Networks are a scalable incarnation of the publish-subscribe paradigm for building asynchronous systems. These take the form of overlays of broker nodes and various routing schemes and filter mechanisms exist that deliver events from publishers to subscribers efficiently on different overlay structures. While a number of principles on which such architectures ride have been well established, and research efforts were concentrated in areas related to information dissemination, efficient routing algorithms, optimal composition and location of filters, optimization of resources use, reconfigurability and fail tolerance, etc., in last years, the attention has turned to exploring non-functional attributes of such systems. Several commercial products (JMS, MSMQ, Corba Event Service, Corba Notification Service) and research projects (Siena, Hermes, Rebecca) implements notification services that considerably differ each other in features and QoS that they provide. The specific nature of the notification services distributed, large scale, big amount of concurrent messages processed, etc., as well as the diversity and complexity of the facets involved in their implementations, impact negatively on the analysis and evaluation of their behavior while they are in execution. This paper proposes a distributed and scalable publish-subscribe broker with support for QoS named QoSEvBroker. The broker leverages on existing mechanisms to reserve resources in the underlying network and on an overlay network of peer-to-peer rendezvous nodes, to automatically select QoS capable paths. By avoiding flooding of either QoS reservations or link-state information, our solution is able to scale with respect to network size and number of reservations. Some experimental results show the validity of our approach.