Download (in spanish only) | ||
---|---|---|
Phd dissertation (2Mb) | Presentation in color(2Mb) | Presentation in B/W(2Mb) |
The demand of Quality of Service (QoS)
and real-time channels on computer networks has recently increased due
to the development of new multimedia applications.
This multimedia transmission
requires a very demanding reservation of network resources. This is due
mainly to the traffic characteristics and the quality of service the applications
require.
This transmission is usually
done using a channel. Channel establishment requires reserving some network
resources (bandwidth and buffers) in order to implement the real-time channel.
This reservation mainly depends on traffic characteristics, but also on
network parameters.
Multimedia traffic usually
has a very variable transmission rate (VBR). That implies that in order
to guarantee the quality of service, the reservation of network resources
is very high. However, these resources are usually misused when transmission
rate is low.
Therefore, the optimisation
of network resources is a key issue. There is a need for methods that obtain
the optimal reservation of network resources which guarantee quality of
service. These methods must be efficient enough in order to be used on-line.
In this dissertation, some
new schemes to save network resources are introduced. The first method
is based on generating a set of points from the stored video with an off-line
analysis of its empirical envelope, and then using these points to efficiently
calculate the optimal bandwidth reservation for a given channel, at channel
establishment time. These points are used to optimise the buffer node.
In both cases, a leaky or token bucket flow specification is used. Finally,
a new scheme for fault detection for highly available real-time channels
is developed. This new scheme provides an efficient resource reservation
and it is based on "suspecting" a failure when the packet delay through
the primary channel is close to the maximum guaranteed delay. Theses schemes
have been proved using MPEG traffic traces.
You can download the papers here: papers