For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
Dynamic Class Assignment for Stream Flows Considering Characteristics of Non-stream Flow Classes
Kenta YASUKAWA Ken-ichi BABA Katsunori YAMAOKA
IEICE TRANSACTIONS on Communications
Publication Date: 2004/11/01
Print ISSN: 0916-8516
Type of Manuscript: PAPER
stream flow, non-stream flow, TCP, UDP, interaction, Diffserv, AF PHB,
Full Text: PDF(732.6KB)>>
In this paper, we term multimedia streaming application traffic "stream flows" and the other usual application traffic "non-stream flows." Many problems occur when both flows are aggregated on a shared link because the different TCP and UDP behaviors cause negative interactions. One way to solve these problems is to isolate stream and non-stream flows to different classes. However, it is difficult to determine the bandwidth allocation for each class and dynamic bandwidth allocation schemes are hard to implement on large scale networks. We therefore propose a dynamic class assignment method that maintains the QoS and that has a higher scalability than dynamic bandwidth allocation schemes. It is workable on Diffserv AF PHB. The outline is as follows. We classify non-stream flows into four classes and dynamically assign stream flows to the classes, taking the conditions and characteristics of the classes into consideration. On assigning classes to stream flows, we map them to a higher drop precedence than non-stream flows not to degrade the QoS of them, based on the assumption that occasional packets being dropped do not create serious problems for them. In this paper, we first discuss our classification of non-stream flows, and present the characteristics of non-stream flows in each class. We then discuss our drop precedence mapping. After this, we propose an algorithm for our method of dynamic class assignment and provide some simulation results where it could provide constant qualities with stream and non-stream flows, adapting to changing traffic.