On hard real-time management information

Maintaining the quality of services for time-critical broadband applications on a high-speed network is an important subject for network management. Since the managed services are time-sensitive, the management applications managing these services must be real-time. This further implies that the real-time management applications must first have real-time management information to correctly take management actions. In this paper, we investigate how to build a monitoring system to support hard real-time management information in the domain of network management. The term “hard real-time” means that there are serious consequences if the management information is older than a specified threshold. The key issue to design such a hard real-time system is to bound the amount of resources that would be needed at the time of monitoring. Three types of resources are under our consideration: communication, computation, and information. This resource bound property motivates us to extend the existing management information protocols (e.g., SNMP or CIMP) to achieve hard real-time monitoring. We describe our experience in building a soft real-time network monitoring system called MELDNET, which uses the shadow object feature provided by the MELDC distributed concurrent object-oriented programming system. Then, we introduce the DONUT model, which is an abstract model for developing real-time scheduling techniques in the domain of hard real-time monitoring. Finally, we discuss the negotiation process between the management applications and the hard real-time monitoring system