Quality of Service in Indirect Communication Systems Filipe ARA ´ UJO Universidade de Lisboa filipius@di.fc.ul.pt Lu´ıs RODRIGUES Universidade de Lisboa ler@di.fc.ul.pt Abstract This paper outlines an architecture that intends to ensure Quality of Service (QoS) to applications us- ing the publisher-subscriber paradigm. The key idea is to subscribe information not only by subject and content but also according to QoS parameters. This must be done with cooperation of all intervenients: publishers, which should tag information they are producing with application-level QoS parameters; subscribers, that do content-based QoS addressing; and message brokers, which use a QoS based net- work infrastructure. 1 Introduction Indirect communication, in particular the publisher- subscriber communication model, is gaining in- creasing acceptance as a useful alternative to di- rect communication models, such as the ones based on remote invocations. The main advantage of this paradigm consists of offering a communication model based on a weak coupling among partici- pants, which do not need to be aware of the location or number of its peers. This simplifies the reconfig- uration of the applications and eases the re-use of components in new applications. A limitation of most existing architectures that support the publisher-subscriber communication, such as for instance, the SIENA (Scalable Inter- net Event Notification Architectures) [4] system, is that they do not offer support for the negotia- tion or enforcement of Quality of Service (QoS) parameters (such as required bandwidth or latency, for instance). This is a significant drawback since QoS features are an important component of mod- ern applications, and its use and support has been widely studied in the context of direct communica- tion [3, 2, 9, 1]. Unfortunately, most of the work on QoS is based on the establishment of channels or connections that reserve the resources required to provide the desired QoS parameters. This paradigm fits perfectly in the direct communication model but does not fit with the decoupling that makes the publish-subscribe model appealing. In the indirect communication model, the applications should not be forced to explicitly set- up channels. Instead, they should remain oblivious to the number and characteristics of the participants involved in the communication and should be con- cerned exclusively on the properties of the informa- tion they are able to publish or subscribe. Therefore, a new system model has to be designed to allow the seamless integration of QoS features in indirect communication systems. This model should: Allow the application to indirectly negotiate QoS parameters, by allowing to express QoS prop- erties has a characterization of the information being produced or subscribed. Delegate on the message broker the task of establishing the required low-level connections, on behalf of publishers and subscribers, based on dy- namic information on the number, location and char- acteristics of producers and consumers and also on the QoS characteristics of the information ex- changed in the system. An interesting feature of the message broker that we are envisioning in our architecture, is that it should be able to autonomously set-up and recon- figure network-level channels. Dynamic adapta- tion is fundamental to optimize the system resources required to convey information from publishers to subscribers with the desired QoS parameters. This is in the lines of the architecture proposed in [6] where the number of IP multicast addresses needed to sup- port a specific subscription pattern is automatically selected by the system. It should be noted that such functionality is more complex than that offered by trading components of direct communication systems [7]. In such systems, a trader is a component that can provide references to server objects that match a set of desired charac- teristics specified by the clients. However, a trader is not responsible for setting up channels between