Adapsjonsmekanismer for tjenestekvalitet i et multimediadatabasesystem

Abstract

Multimedia data are characterized by large, complex values with temporal dependencies. Such data are used by distributed multimedia applications offering long-lived multimedia presentations in distributed environments. An end-client may choose to specify quality requirements for a multimedia presentation, using quality of service (QoS) parameters known from traditional communication technology. The client's demands and requirements should be satisfied throughout the entire presentation, thus QoS management is required. These management tasks may be performed by software components residing at each functional component in a distributed multimedia system, for example the network or database system. We refer to such a software component as a QoS manager, who, among other things, is responsible for adjusting the level of QoS prior to and during a multimedia presentation. Whenever the user's requirements regarding QoS are not satisfied, the adaptation mechanism is responsible for bringing the QoS level within the accepted boundaries.

We focus on adaptation mechanisms for QoS in a multimedia database management system (MMDBMS) offering services to applications for distance learning called Learning-on-Demand (LoD). LoD offers interactive services and a high degree of user interactions.

In the OMODIS (Object-Oriented Modeling and Database Support for Distributed Multimedia Systems) project at the Department of Informatics at the University of Oslo, a framework for QoS management has been developed. The framework defines messages for interaction between QoS managers in a dynamic, reconfigurable QoS management hierarchy. Each manager consists of 7 internal agents, one being a QoS adaptation agent. In our work we propose messages for interaction inside a QoS manager connected to a MMDBMS and a QoS management hierarchy. To decide the tasks of a manager for a MMDBMS we identify a set of events that may occur in a MMDBMS supporting LoD applications. These events are used to design protocols for internal communication between the agents of a QoS manager for a MMDBMS.

The internal messages are realized through formal specifications in Promela (protocol meta language). The formal specifications are used to simulate and validate the behavior of the internal message passing. Simulations and conformance testing are used to show that the protocols as specified, are correct with respect to a set of correctness criteria, and that the protocols support the agents in performing their QoS management tasks.

Temporal claims and correctness requirements are used during validation. The validation results show that the protocols for internal communication meet all correctness criteria. In particular, the protocols are free of deadlocks and livelocks. All protocol executions end in legal end states, and they all follow our definition of legal sessions in respect to the session states defined in the OMODIS QoS framework.

In addition, we propose an adaptation model for the QoS adaptation agent of the QoS manager. The model defines behavior for local QoS adaptation prior to and during a presentation. Our adaptation model is based on QoS monitoring results of the local component. Thus, it is capable of discovering QoS violations caused by the MMDBMS at an earlier stage than adaptation models depending exclusively upon monitoring results from the client side. Problems at the client side will not arise until an end-to-end QoS violation has occurred.