| dc.description.abstract | Future high speed network, such as Broadband ISDN, will be required to carry the traffic generated by a wide range of services. These services will have very diverse traffic flow characteristics and performance requirements. Some of the many factors that have led to this diversified nature of high speed network traffic are:
Services integration
Broad range of compression schemes
Broad range of applications
Many of the high speed network sources initially generate a continuous stream of high speed data. To reduce the bandwidth needed, sources always employ various compression techniques. Thus, sources belonging to the same class itself would require different performance guarantees because they employ different compression techniques.
In some cases, applications themselves contribute to the diversity of high speed network traffic. For instance, consider a News-on-Demand (NoD) application which generates three temporally synchronized streams:
Low bit rate video of the news reader
High bit rate video of the actual news
Low bit rate data of general announcements
So the worst case is a multimedia call where there is application-induced heterogeneity within a single call.
In order to provide guaranteed performance levels and utilize the network resources simultaneously, preventive control approaches such as Call Admission Controls (CAC) are used. In this approach, traffic sources describe their behavior using some appropriate parameters and also their performance requirements to the network before establishing a connection.
From this traffic description and the current status of the network, the network decides whether to provide a service for it or not. The common drawback of almost all CACs that have so far appeared in the literature is that they presume all sources as random traffic sources, i.e., they expect all sources to describe their traffic with some set of statistical parameters. But in practice, not all applications generate random traffic flow. In fact, all applications associated with real-time play-out of pre-recorded and pre-orchestrated multimedia (MM) data are deterministic traffic sources. There are many applications dealing with pre-recorded data such as:
Multimedia Archive Retrieval System
Video-on-Demand
News-on-Demand
In these cases, sources generate very diverse but deterministic traffic flow, because the traffic of a pre-recorded source is completely known a priori.
Based on this idea, this dissertation proposes a new hierarchy of classification of traffic and identifies a new traffic class called "deterministic class".
In this dissertation, we proposed a new mathematical traffic model to represent the traffic generated by a pre-recorded multimedia source, based on weighted sum of shifted unit functions.
Based on this mathematical model, we proposed a Call Admission Control (CAC) exclusively for the deterministic traffic class, which provides deterministic guaranteed QoS using minimum buffers. This CAC does not take into account buffer availability at each intermediate node while making a decision on a connection’s admission. In other words, it assumes infinite buffers at each intermediate node.
We also provide a method to integrate our CAC with other CACs meant for statistical traffic. Additionally, we discussed the behavior of this CAC when finite buffers are available at each intermediate node.
We studied the performance of this CAC by simulating it extensively in a partially connected arbitrary topology network in both situations:
Finite buffer availability
Infinite buffer availability
We studied the effect of:
Call arrival rate
Link capacity
on performance parameters such as:
Utilization
Blocking Probability
Average Starting Delay
Results show that superior performance can be achieved over statistical CACs, if separate traffic controls are applied for this deterministic traffic. | |
