Reserving for future clients in a multipoint application-why and how?

  • Authors:
  • P. Moghe;I. Rubin

  • Affiliations:
  • California Univ., Los Angeles, CA;-

  • Venue:
  • IEEE Journal on Selected Areas in Communications
  • Year:
  • 2006

Quantified Score

Hi-index 0.07

Visualization

Abstract

As applications become more sophisticated, we believe that the problem of guaranteeing their service commitment will get much harder. Currently applications requiring real-time media quality of service (QoS) are set up using a flow-reservation model. In this model, the network admits each flow (or connection) within an application separately using RSVP, asynchronous transfer mode (ATM) signaling, or some other mechanism. This model is restrictive because it cannot inherently deliver guarantees on future client additions to an ongoing application. We claim that such guarantees are going to be required to build sophisticated multipoint and multiparty applications that involve a dynamic number of clients and a real-time sharing between clients. Without such guarantees, existing clients within an application can be subjected to context disruption or unreliable communication. To reduce this form of degradation, we propose the idea of application reservation where resources for an entire application are reserved at application setup. This includes resources for the clients expected to add to the application in future. Application reservation can be implemented as an overlay to existing flow-based reservations. We present the architectural features of such a framework. As part of the service interface, applications declare space-time dynamics of future client arrivals in addition to traditional descriptors. A new class of QoS measures that quantifies the type of degradation mentioned above is also conveyed. The application manager checks to see if the application can be admitted. After the application is admitted, the manager monitors clients as they arrive, to ascertain if they conform to the declared service interface parameters. Conforming clients have resources reserved for them, so they can be simply added to the application. We review these architectural components and address the nontrivial problem of managing the nonconforming clients