
Open access
Date
2002-05Type
- Report
ETH Bibliography
yes
Altmetrics
Abstract
We consider the problem of scalable call admission control (CAC) in a DiffServ network and propose a new CAC scheme. Initially, a certain fraction of the bandwidth of the network is reserved for admission-controlled traffic, and this reservation is carried out according to the hose model, a model that had previously been proposed for VPN provisioning. With this architecture, admission control can be carried out at edge routers without any need for global coordination. Following the DiffServ philosophy, routers in the core of the network are not involved in CAC or signaling and do not need to be aware of individual flows. This ensures scalability while offering deterministic bandwidth guarantees. Assuming that different connections create different amounts of profit for the provider, we consider CAC algorithms that try to maximize the obtained profit. In the on-line version of the problem, we observe that the algorithm by Awerbuch, Azar and Plotkin can be implemented using local control and achieves a logarithmic competitive ratio. In the off-line version, where all connection requests are known to the algorithm, we give efficient approximation algorithms achieving good approximation ratios. These algorithms are useful in a scenario with advance reservations. We also give a detailed specification of a possible implementation of the proposed CAC architecture, and present results of high-level simulations showing the relative performance of different CAC algorithms and the influence of a user model where users may sometimes be willing to accept an increased price for their connection. Show more
Permanent link
https://doi.org/10.3929/ethz-a-004338476Publication status
publishedJournal / series
TIK ReportVolume
Publisher
ETH Zurich, Computer Engineering and Networks LaboratoryEdition / version
Version 1Organisational unit
02640 - Inst. f. Technische Informatik und Komm. / Computer Eng. and Networks Lab.
More
Show all metadata
ETH Bibliography
yes
Altmetrics