Design and Implementation of Telecommunication Access Network

Expansion Planning System

Marijan Valković Branko Lukić Dragan Čišić

Damir Medved2
[email protected] [email protected] [email protected]
[email protected]
TeleConsult d.o.o. - Rijeka, Croatian Telecomm Faculty of Maritime Studies
G. Krkleca 7 Telecommunication Center Rijeka University of Rijeka
HR-51000 Rijeka, CROATIA HR-51000 Rijeka, CROATIA HR-51000 Rijeka, CROATIA

Abstract The main purpose of this paper is not to discuss the

detailed modelling issues of the local access network
Technological innovations and growing consumer expansion problems, rather our emphasis is to show
demand have led to a variety of design and expansion design and implementation details of the algorithms
problems in telecommunication networks. In in the optimization system called LANEP that will
particular, local access networks have received a lot hopefully become a standard building block of the
of attention, since they account for approximately company decision support system.
60% of total investments in communication facilities.
Hence even marginal improvements in the access
network design can lead to substantial savings. 2. Background Information
In this paper we describe software tool that handles The first author has gained substantial experience in
so called "Local Access Network Expansion the development of telecommunication network
Problem" (LANEP) where growing demand can be planning tools while developing Windows version of
satisfied by expanding cable capacities and/ or ITU PLANITU software. Program algorithms were
installing concentrators in the network. The implemented as FORTRAN subroutines with huge
problem is known to be NP-hard combinatorial named common block as data interchange vehicle.
optimization problem. Implemented dynamic The GUI was implemented with C++ and MFC
programming algorithm is due to Flippo et al. [4] objects.
and exploits tree structure to solve the problem to Bare bones FORTRAN programming is still suitable
optimality. only for console applications due to favorable
numerical performance and advanced debugging
1. Introduction facilities (array bounds checking and program error
traceback) of contemporary FORTRAN compilers.
Design and expansion problems arise frequently in
Local Access Networks (LANs), which typically Full GUI, user-friendly software tools that are
have a tree structure. If the number of users in the expected by most planners, are commonly developed
LAN is small, a star configuration in which all users with so-call Visual tools as Visual C++ or Visual
are directly connected to the central component is Basic in an object oriented programming (OOP)
often employed. As the number of users increases, fashion. While the integration of both worlds (old
however, costs can be reduced by introducing a more style FORTRAN plus OOP C++ code) is possible (as
general tree structure, and by installing concentrators exemplified with PLANITU) its development cycle
in the network. Together with the central component, is longer and program maintenance much more
these concentrators could form a backbone network difficult than in the single-language programming
(ring structure) to increase reliability, or they can also case. For example hybrid programming environment
simply be connected to the central component via a necessitates programming expertise in two quite
direct line. To optimize these networks, a two-phase distinct kinds of programming art and two absolutely
procedure is often proposed. compatible compilers (often debugging symbols are
not compatible). It is well known fact that C++
In the first phase the user nodes are partitioned into compiler technology is developing at much faster
regions, while in the second phase concentrators are pace than FORTRAN compiler technology, so
placed in each of these regions to meet traffic commercial C++ compilers are updated more
demand between user nodes, see [6]. An overview of frequently creating additional problems with code
various capacity expansion problems can be found in maintenance and program updates.
[1] and [5].
Stimulated with the above mentioned facts and Condition 1 reflects guidelines of network planners
confronted with the limited workforce and time who, given the current relative costs of cable
frame for LANEP development it was decided to expansion and concentrator installation, consider
switch completely to OOP paradigm and utilize multiple levels of concentration to be uneconomical.
"Library of Efficient Data types and Algorithms" - Conditions 2 and 3 are enforced to ensure operational
LEDA [8] for problem modeling. convenience of maintenance and repair.
On the other hand input data for access network
planning are numerous and are rapidly changing, so 3.1. Cost Structure
an advanced GIS system is a prerequisite.
Algorithm can handle general cost structures for
In our case planning data was extracted from the GIS cable expansion and concentrator installation costs,
system mainly supporting network operation and which are node and edge dependent.
maintenance. Due to the fact that GIS system was not
initially designed for network planning tasks, certain For every edge e an existing edge capacity be and
adaptation was necessary. edge dependent fixed Fe (trench and duct cost) and
variable ce (cable and installation costs) cost are
defined with real valued edge expansion cost
3. Problem Description function Le .
In LANEP a tree is given in which a number of user
nodes are connected to the switching center in the Le (le )  Fe  (le  be )  ce for le  be ;
root of the tree. Each user node typically represents a
and zero otherwise
collection of individual users connected by an
underlying network. Communication between user where le is edge traffic load.
nodes of this and other LANs is accomplished
Every concentrator of type t has a given capacity bt
through the switching center. Therefore, instead of
and node dependent fixed (Fvt) and variable (cvt).
using traffic demand between pairs of nodes, we can
installation costs
assume that each user node has a demand that must
be routed to the central switching center. This may be For every node v a traffic demand rv and a real valued
accomplished in two ways, either by routing the cost function Kv is given, where Kv(kv) specifies the
demand to the switching center via its unique path in concentrator costs for the load of kv processed in
the tree, or by routing it to one of the concentrators node v.

Kv(kv)  min t Fvt  k v  cvt | bt  kv  for kv >0;

that are (to be) installed in the network. A
concentrator compresses all incoming low frequency
signals (demand) into one outgoing high frequency
(or optical) signal, which is then routed to the and zero otherwise
switching center. It is assumed that this outgoing
signal either requires negligible capacity in the Fixed concentrator cost Fvt consists of land and
network, or is routed to the switching center via a building costs, electric power feed fee and various
dedicated line not belonging to the network. The taxes (building permit, etc.) related to specific node
costs of constructing such dedicated lines are v. Variable cost component c vt consist of concentrator
included in the installation costs of the concentrators hardware cost and transmission cost for the high-
involved. In practice, a large variety of electronic speed link between concentrator and the host in the
devices are available to compress signals. For the root node.
problem at hand, we can simply treat them as
concentrators with different capacities and
operational costs. 3.2. Actual Costs
The routing of demand in the LAN is restricted to the Concentrator and cable expansion costs are
following conditions. commonly modeled as piecewise-linear, concave
function, see [2]. In reality, concentrator or cable
1. Single level concentration: demand is expansion cost is a step function because
concentrated at most once before reaching the concentrator modules are available in discrete units.
switching center in the root of the tree; However due to rapid technology changes it is
2. Nonbifurcated routing: for each user node its difficult to predict cost developments in the future, so
entire demand is processed by a single especially for long-term planning adopted model is
concentrator (possibly at the root); more than adequate.
3. Contiguity condition: if a node v homes on a Cost figures for four types of concentrators (256,
concentrator in node w, then all nodes on the 512, 1024 and 2048 lines) and three types of
path from v to w home on w. concentrator housings (cabinet, container and small
building) were analyzed. To limit the number of
combinations and emulate as far as possible real pending demand and upgrade the OSP with new duct
world planning problem, input data for two smaller routes and new buried cables of adequate capacity.
concentrators in cabinet and two larger concentrators
Kozala service area belongs to the metropolitan
in buildings were actually prepared for LANEP, see
telecommunication network of Rijeka characterized
Table 1 and Table 2.
with high subscriber density and well defined outside
Table 1 Edge expansion costs. plant on the southern part, and almost rural-like
characteristics on the northern part of the investigated
Edge Expansion Cost Type EUR/m area, see Figure 1.
Fixed cost 40,00
Variable cost EUR/m/pair Therefore, substantial investments are necessary in
10 pairs 0,1200 the low-density mainly aerial-cable fed northern part
of the service area.
50 pairs 0,0520
100 pairs 0,0425
200 pairs 0,0370
500 pairs 0,0350
1000 pairs 0,0320
Table 2. Concentrator expansion costs.
Concentrator Cost Type EUR
Fixed cost
256 and 512 capacity
concentrators in cabinet.
land cost (25m2) 550,00
building costs (foundation, etc.) 550,00
power feed fee 1100,00 Figure 1 Kozala exchange area access network
 2200,00 topology in the GIS system.
Up to 2048 capacity
Currently there are 8692 connected subscribers in the
concentrators in container.
2
land cost (40m ) 1000,00
Kozala area out of which 2543 are duplex
building costs (foundation, etc.) 1600,00 subscribers. Within the area there are 6236
building fee 1400,00 apartments in block houses and 1121 apartments in
container cost 4100,00 family houses. Expressed demand (waiting list)
power feed fee 1100,00 currently accounts for 541 new telecommunication
 9200,00 lines.
Up to 2048 capacity GIS database currently holds site and schematic cable
concentrators in building. diagrams, duct and manhole drawings as well as OSP
land cost (100m2) 2200,00 (DP, CCP) and switching equipment location and
building costs (24m2) 36000,00 capacity. Georeferenced building locations with
building fee 1400,00 forecasting data attached are available for network
power feed fee 1100,00
service forecasting. Up to now primary purpose of
 40700,00 the GIS in TKC Rijeka was to support network
Variable cost EUR/line operations and maintenance. Practically all
Concentrator in cabinet departments use this database, primarily for real
256 185,00 world visualization of telecom infrastructure,
512 185,00 localization of demand for new lines, equipment and
cable failure troubleshooting.
Concentrator in
container or building. When the GIS database was created, the primary goal
1024 100,00 was to establish connection between alphanumerical
2048 90,00 data in SQL DBMS and georeferenced information
contained in numerous AutoCAD schematic and site
drawings. In the future, special attention will be paid
4. Network Topology and Customer
to adopt data acquisition and preparation procedures
Database that will emphasize the importance of network
Exchange area Kozala was selected for the trial planning data in addition to O&M related data, and
evaluation of LANEP because it is facing a major this is the direct consequence of the development and
overhaul due to the need to eliminate duplex lines implementation of LANEP for the Kozala network.
from the network, provide new capacity for the
4.1. GIS Data Extracted and Used were selected as graph nodes. Duct and/or cable
routes with aggregated cable capacities constitute
Fort the first exercise, existing terminal boxes with graph edges.
typical capacity of 10, 20, or 40 subscriber lines,

Figure 2 LANEP main menu and graph editor window with the structure of the Kozala network
prior to optimization.

5. Overall System Design 5.1. User Interface

The LANEP system is implemented in MS Visual User friendly interface with simple to handle pull-
C++ 5.0 and LEDA. Main optimization algorithm down menus, see Figure 2, simplifies planer
expects network topology in a tree structure with interaction with LANEP. Besides that, mouse
nodes numbered in DFS order with the root pointer and mouse buttons in combination with Ctrl
numbered 0. The edges are also numbered in DSF and Alt keys handle basic system interface
order such that edge v is the edge between node v functions, as zooming and attribute (color, label
and its unique parent. LANEP utility functions can type, line/ node symbol style, etc.) set-up for graph
transform arbitrary network graph into DFS form. nodes and edges.
LANEP Parameters sub-menu is used to set the
number of different concentrator types,
 Once the graph is read, it can be interactively
edited prior to network optimization with LANEP.
Graph construction for every new project is always
a tedious task even in the case of data exported
from a reliable GIS. Often network topology is not
and set their size. required in the GIS, so what seems to be a perfect
connected graph on workstation screen, might in
reality be unconnected graph. The first attempt of
graph construction for the Kozala network, resulted
in a graph with 8 connected components. In such
case DFS algorithm fails to label nodes (in DFS
order), in graph components that are not connected
Prepare DFS Format for LANEP sub-menu to the graph component with the root node in it, see
command: Figure 3.
 Creates Minimum Spanning Tree (MST) of the Figure 4 and Figure 5 examplify interactive graph
network graph alredy present in the LANEP editing and provide insight into LANEP data
graph editor. requirements.
 Prompts the planer to select the root node
(exchange location).
 Subsequently, runs Depth First Search (DFS)
algorithm and topologicaly sorts nodes and
edges starting from the root node.
 Finally, writes-out LANEP formatted project
file acording to the format introduced in [9].

5.2. New Network Graph Construction

and Editing
New LANEP graph can be constructed from
scratch using embedded graph editor or can be
input from ASCII (nodes and links) files exported
from the GIS. While the first method of graph
construction is suitable for small graphs, the second
approach is the only practical solution for large
networks with few hundred nodes and edges.
Figure 4 Interactive edge editing.

Figure 3 Missing link has caused the DFS

algorithm to fail to label a cluster of
unconnected nodes (light colored nodes).

Figure 5 Interactive node editing

6. Computational Results one concentrator type of 1024 lines, see Figure 6
and Figure 7.
Various scenarios for the Kozala network were
prepared and run with the LANEP. In particular Running times of the algorithm on the Kozala
two scenarios were found distinct enough to be network whose graph has 547 nodes and 546 edges,
published here, investigation (i) for two on P200 computer with 64 Mbytes RAM, were
concentrators types of 512 and 2048 lines 54.739 sec and 17.625 sec for the investigation (i)
optimized simultaneously and investigation (ii) for and (ii) respectively.

Figure 6 Optimal concentrator locations and capacity for the concentrator types 512 and 2048 lines.

1. implementation of an efficient and proven

7. Further Work LANEP algorithm,
Tighter integration of LANEP with the GIS is a 2. elegant library of efficient data types and
natural next step that will provide more efficient algorithms with integrated graph object
planning environment. In addition to that, new enabling fast problem modeling and
forecasting modules will automate now tedious visualization,
manual clustering of demand for telecom services
and provide direct input to LANEP. 3. up-to-date planning data from the modern GIS,
has enabled a small team of professionals to
8. Summary and Conclusions achieve valuable results in a short available time.
The three pillars of the project;
 Figure 7 Optimal concentrator locations and capacity for the concentrator type 1024 lines.

Algorithms implemented in a productive [3] G. Cho and D.X. Shaw. "Limited column
environment come alive quickly producing very generation for local access telecommunication
useful results; and this is an extremely important network design - formulations, algorithms, and
step in a long way to self-contained software tool implementation", December 1994.
development.
[4] O.E. Flippo, A. W.J. Kolen, A.M.C.A. Koster,
R.L.M.J. van del Leensel, "A Dynamic
References Programming Algorithm for the Local Access
Network Expansion Problem", Universiteit
[1] A. Balakrishnan, T.L. Magnanti, A. Shulman, Maastricht, Research memorandum 96027,
and R.T. Wong. "Models for planning capacity September, 1996.
expansion in local access telecommunication
networks". Annals of Operations Research, 33:239- [5] B. Gavish. "Topological design of
284, 1991. telecommunication networks: Local access design
networks". Annals of Operations Research, 33:17-
[2] A. Balakrishnan, T.L. Magnanti, R.T. Wong, 71, 1991.
"A decomposition Algorithm for Local Access
Telecommunication Network Expansion Planning", [6] L. Gouveia and J. Paixao. "Dynamic
Operations Research, Vol. 43, No. 1, January- programming based heuristics for the topological
February 1995.
design of local access networks". Annals of
Operations Research, 33:305-327, 1991.
[7] C. Jack, S.R. Kai, and A. Shulman, "Design and
implementation of an interactive optimization
system for telephone network planning",
Operations research, 40:14-25, 1992.
[8] K. Mehlhorn, S. Nähaer, M. Seel, C. Uhrig,
"The LEDA User Manual Version 3.7", Max-
Planck-Institut für Informatik, 66123 Saarbrücken,
Germany.
[9] D.X. Shaw, "Limited Column Generation
Technique for Several Telecommunication
Network Design Problems", December 1994.