|
|
|
Project Highlights
- Self-Scheduling in High Speed Data Networks
Modern telecommunications networks are expected to transport
diverse sources of traffic such as voice, data, video on demand
and multimedia. Designing and managing such a single network
that can perform as well as the public switched telephone network
can be very challenging. Each traffic type requires different amount
of network resources (buffer space and bandwidth), and its flow
through the network is affected by network control functions such
as queueing mechanism, bandwidth allocation, and routing protocols.
Efficient and easily implementable network management
algorithms are required to properly guarantee Quality of service
(QoS) to all types of traffic and at the same time reduce congestion,
maximize network utilization and ensure that the network is stable.
The self-x framework we introduce here for the basic network management
functions is part of the vision of a self-x network that can manage
its functions itself without requiring human interactions. It is
expected that such a network will transport diverse sources of
traffic such as voice, data and video on demand, while guaranteeing
individual and specific QoS to each user. It
is also expected that such a network is self-installing, self-learning,
self-sizing and self-healing. In order to achieve these elf-aspect
goals, the work should possess: (i) the self-knowledge which is typically
derived from accurate on-line measurements and current network conditions,
and (ii) the self-learning abilities to adapt to changes in traffic
and network conditions.
The self-x framework we propose here consists of three modules
and addresses separately three functions namely, traffic prediction,
traffic scheduling and connection admission control. In general,
Connection Admission Control (CAC) can be defined as the procedure
of deciding whether or not to accept a new connection. One of
the fundamental aspects of CAC is to evaluate the impact of a new
connection on the current traffic load. This usually involves determining
the resources (bandwidth and buffer space) needed for a new connection
with its specific (QoS)
in the presence of existing connections. If the traffic generated
by each connection is a deterministic function of the time,
then the CAC procedure could consist of simulating off-line
the traffic forwarding mechanism and in estimating the level
of service provided to each. In practice, most of the
time, the traffic is highly variable and can
only be expected to provide its statistical characteristics. The
notion of the effective bandwidth of a traffic source has been used
recently to describe the effective resource requirements. This procedure
is based on large deviation theory and depends strongly on the statistical
properties of the traffic sources. Most of the methods require
accurate statistical characteristics of the sources that may not
always be feasible to obtain in practice. Besides, sources are
assumed to be active indefinitely. In order to design
an adaptive CAC, we introduce the notion of {\it Virtual Line Rate
Required (VLRR)}, an estimator of bandwidth. For a set
of sources, we use the virtual traffic, as opposed to the actual
traffic. The virtual traffic is either the predicted traffic
(that is calculated by the Fuzzy Traffic Predictor), or the
approximated traffic. For a given mean rate $m$ and
a peak rate $P$, the approximated traffic is a sequence of random
numbers
between $0$ and $P$, generated according to a uniform
distribution such that the mean of the
approximated traffic is close to $m$. The virtual
traffic is then forwarded to a virtual
channel using the Fuzzy Traffic Scheduler whose capacity
can be fixed.
|
|