| Summary: | 碩士 === 國立臺灣大學 === 電機工程學系 === 86 === Due to the fast development of high speed asynchronous transfer
mode (ATM) networks, the volume of carried traffic should
increase greatly in the future. The steady increase in optical
transmission bandwidth suggests that even a single link failure
will seriously impact many service users.
Thus, network survivability is a key concern for future ATM-
based networks. When a failure occurs, a fast restoration
protocol is needed to accelerate the restoration speed and
recovered affected bandwidth as much as possible (restoration
ratio). However, the success of fast restoration significantly
depends on how traffic is distributed and spare capacity is
dimensioned over the network when a network failure occurs. In
other words, the restoration process alone is not enough to
achieve a high level of net
In this thesis, we study and design a centralized asynchronous VP configuration
system for ATM networks. There are four stages of our research:
(1) configuration problem description and mathematical
formulations, (2) solution algorithm design, (3) VP
distribution architecture and VP setting protocol design, and
(4) reference
implementation.
In the first stage, we describe the VP configuration problem of an ATM
Network. There are two sub-problems: regular VP assignment and backup VP
assignment. For regular VP assignment problem, end-to-end QoS (Quality of
Service) requirements at the cell-level, such as Cell Delay
Variation, Cell Delay and Cell Loss Rate, are
considered by the concept of contract region where each traffic
class is given an "Equivalent Capacity" over the links
traversed by this traffic. Call-level QoS requirements such as
call blocking rate are set to be bounded for every S-D (Source-
Destination) pair of each traffic type, and call setup
times are also bounded by limiting hop number traversed by each VP.
Signaling capacity is also modeled as
a major constraint. Based on the QoS requirements and constraint
consideration, regular VPs are configured to minimize the maximum single
unidirectional link load ratio over the network for the sake of network
survivability: just do not put all the eggs in one basket.
In order to
increase the network survivability, we adopt preplanned backup VP routes
to deal with the interrupted services over regular VP routes.
Backup VP assignment tries to
maximize average restoration ratio for all link
failures based on calculated regular VP configuration. In this
thesis, only single bi-directional link failures are taken into account and
we assume that each regular VP has at most one backup VP route which goes
through different physical links. A model is also
proposed to compute restoration ratio due to each link failure occurrence.
The Erlang's loss formula based on M/M/c/c queue model is
adopted to calculate the call blocking probability
for each S-D pair. Node signaling load is approximately expressed by call
request and release messages passing each node. We assume that
an incoming call first attempts the first route in the possible
set
of routes according to hop number limitation. If the call is blocked, the
second route is attempted, and so on. Besides, only direct VPs from source
node to destination are considered in our formulation instead of concatenated
VPs.
We propose a Regular VP Assignment Algorithm (RVPAA) to deal with regular VP
configuration problem. The call-level and cell-level QoS of multiple traffic
classes can be guaranteed in RVP configuration. Simulation results based on
NTU ATM backbone network demonstrate the step by step
evaluation of our RVPAA. We also compare our solutions by our
RVPAA with optimal solution by an exhaustive algorithm. Our
solutions stays close to the optimal
Maximum Link Load Ratio (MLLR) solution with average 17\%
probability to find the
optimal solution and with 100\% restoration ratio. In addition,
computational time of our RVPAA is much smaller than that of
exhaustive search. We also propose
Backup VP Assignment Algorithm (BVPAA) to handle the backup VP
configuration problem based on the calculated regular VP
configuration from RVPAA. Preplanned backup VP routes are
configured in order to increase the restoration time and
restoration ratio. Key ideas of baseline solution algorithms
for RVPAA and BVPAA are random initialization, constraint
violation reduction and maximum index value reduction. The
first two steps are for feasibility and the last step is for
optimality.
Simulation results based on Taiwan testbed network demonstrate that average
restoration ratio based on our RVPAA and BVPAA can be enhanced
when single bi-directional link failure occurs.
A VP setting protocol based on SNMP configuration is developed
to realize the graceful VP distribution mechanism. We proposed
a Manager VP setting protocol at the manager station and a
Agent VP setting protocol at each agent. Before downloading
configuration data to current VP transport network, several
procedures must be taken into consideration in order not to
interrupt connections currently in service.
Finally, we conduct a reference implementation of our
centralized asynchronous VP configuration system. There are two
Fore ATM switches in this reference implementation. The
centralized asynchronous VP configuration system is designed at
the network management station under SUN-OS 5.5.1. The manager
process is designed to update the ATM inventory database and
distribute the VP configuration data to each network elements.
Besides, the agent process receives the information from
manager process, then update t
|
|---|
