APPLICATION OF NETWORKS PLANNING TOOLS FOR INFORMATION CONTENT ALLOCATION

1
APPLICATION OF NETWORKS PLANNING TOOLS FOR
INFORMATION CONTENT ALLOCATION
LONIIS

• A.Koucheryavy, V.Lokhmotko, Z.Revelova,
  A.Paramonov
• LONIIS, Russia, 196128, Warshawskaya, 11, St.
  Petersburg,
• phone [phone], fax [phone]
• E-mail priem_at_loniis.org

2
CONTENTS
LONIIS

• Introduction
• Generalized formulation
• Client ability evaluation
• Algorithm of content allocation
• Numerical example
• Summary

1
3
Abstract
LONIIS

• In this paper the regional problems development
  of the Russian segment Internet are considered.
  The international Internet experience is
  analyzed. The problem of optimum information
  content allocation has been defined. For solution
  of this problem it has been offered the use of
  the telecommunication networks planning tools
  which permit to obtain additional quantities of
  channels, switching and hard disk memory
  resources of a network by means of redistribution
  of information flows and transmission capacities
  on the networks graph. To obtain an allocation of
  copies specification it is offered to use
  clusterization algorithm and for copies
  quantities a method of one-dimensional
  searching. An example on quantitative method is
  provided.

2
4
1. INTRODUCTION
LONIIS
By today, the capabilities of Internet providers
are not limited by global network accessibility,
development and Internet resources maintenance,
development and arrangement of advertising, but
are much wider as observed below

• active consolidation and Internet providers
  attachment which are based on vertical
  integration, synergetic effects and additional
  benefits expressed in cost reduction and
  customers base expansion
• increase on the demand of mobile Internet,
  high-speed and broadband access with a
  simultaneous decrease of a demand on switched
  access
• nomenclature extension and service quality
  improvement
• adoption of new business models by Internet
  providers including
• creation of portals and content-projects,
  provision of services on information for
  users Internet sites, development of Internet
  data computer management systems
• provision of complex services for organization
  and support of an effective processing of
  fundamental WEB business-tasks
• WEB based decisions for interactions between
  companies (e.g. trade, information exchange
  etc.), and also between companies and customers
  (customers support on-line service, Internet
  shops etc.).

3
5
LONIIS
Internet application problems in Russia
The Russian Internetization yet has not
achieved a global level. In the terms of "life
cycle production graph" it meets to implantation
and initial growth phases . Up to date, the
benefits level of an average Russian operator due
to Internet services provision corresponds a 5
part of an overall benefits. The problem of the
Russian Internet segment expansion is wrapped
by a set of application problems, which are the
following

• determination of traffic routes for each type of
  service
• providing maximum enclosure of Internet traffic
  within the region
• selection of the best strategies of Internets
  services provision, e.g. Internet services with
  required QoS
• calculation of an additional amount of channel
  and commutative resources necessary to expand a
  packet switched network
• determination of optimum quantity and most
  preferable variants of server allocation.
• The last problem is mainly considered
  in the given paper.

4
6
2.GENERALIZED FORMULATION
LONIIS
The basis of a creation (expansion) of an
existing regional Internet structure forces us to
consider it as a symbiosis of information
networking system and packet switching network.
Further, we take into consideration the
dominating Operator commercial concerns and a
client ability as well. It supposed that we know

• a topology of an existing packet switching
  network and overall parameters of equipment
• allocation of Internet users and servers per
  administrative regions of a country
• forecast on Internet users quantity and Internet
  traffic, structure and grade of a services to be
  provided.

5
7

LONIIS
Structure of a Network
The structure of a packet network is represented
by the distributed graph G (E, V, A) with a set
of top points ?, a set of edges V and a topology
A. Lets consider, that an existing background
traffic ? f is defined already and in particular
case it is equal to zero, and both a forecast of
request-traffic ? r is already known and its
allocation on the regions is known as well. The
conditions are as following

• each node may be equipped or not with a server,
  i.e. in general, a number of servers Ns is not
  equal to number of clusters Nu
• servers contain copies of objects (programs,
  data, images etc.)
• network can operate in excess and deficit of
  transmission capacity conditions
• any server can operate as an initial server,
  proxy, gateway or tunnel, changing its behavior
  according to request type
• application of protocols ??TP, FTP, Gopher, SMTP
  etc. as generalized communications protocols
  acting between users agents (browsers, editorial
  or another automatic tools) and proxies /
  gateways or another Internets tools
• centripetal (directed to the nearest local
  server) behavior of Internet traffic ?r . In
  absence of necessary data domain or information,
  the request should be transmitted to other
  servers of this region, or to "external" servers
  of other regions, and abroad as well
• multi-variant access.

6
8

LONIIS
Required
To find an optimum number Ns of local servers,
most preferential variant of their arrangement by
switching nodes and values of additional network
equipment which maximize parameters of Operator
commercial activity (pure discounted benefits,
investments profitability, capital efficiency
etc.) on a given planning level, or minimize

• a cost of information storage, transmission,
  upgrade and retrieval
• volumes of stored and transmitted information
• a cost of necessary additional channels and
  switching operational resources for expansion of
  an existing Internet network segment
• a degree of servers and communication channels
  load uniformity
• a value of duplicated information.

The solution provided in this paper satisfies
international standards and design constraints
including QoS, capacity, availability, degree of
information requirements sufficing,
fault-tolerance, transmission rate,
survivability, readiness, security etc.
7
9

LONIIS
3. CLIENT ABILITY EVALUATION
The forecast of client base increase is based on
calculation of possibility q(?) that the customer
with the benefit from ? up to (? d?) becomes an
Internet user, or a similar possibility F(?). The
metric 100q(?) is interpreted as part of
population (rather 100 persons) with the monthly
benefit from ? up to (? d?), equipped with
Internet, and F(?) is a number of persons having
the benefit ? and more. According to the Pareto
law, confirmed in many countries and after some
centuries, the number of persons having the
income ? also is more defined by ratio F(?) A?
-? , ( A and ? distribution parameters, ? 1,5).
According to the formula F(?) the values in Table
1 are calculated appropriate to Operator having a
low-benefit urban users and a large-scale urban
areas. Integrated overall economics of such
Operator (on the extremity of 2001) consist of
8
10
LONIIS

• overall telephone subscribers quantity is 300000
• density of telephonization is 24 telephones per
  100 inhabitants
• overall Internet users number (including 20
  corporate users) is 13000
• average level of population revenue on
  Archangelsk region is 3R ruble /month (R is a
  statistical parameter)
• average revenue of Internet Operator is R
  ruble/year per user
• part of the average Internet user expenditures
  from an overall revenue is about 3
• telephonization level is 24
• internetization level is 1 .

9
11
Model of traffic
LONIIS
Table 1
10
12
As it is shown in table 1
LONIIS

• for Pareto asymmetric distribution characterized
  by displacement of a mode to the side of low
  income, only about 9 of population has the
  monthly income 3 000 rubles and more
• the existing level (1 ) of internetization meets
  a sparse interlayer of the population from 3-rd
  and 4-th groups with monthly earnings not less
  than 12 000 rubles. To number of the Internet
  users can be treated also persons with ? lt 12 000
  ruble/month, but already with probability
  distinct from unit
• rate of growth of integrated function F(?) is so
  that at magnifying of an average level of the
  population incomes at 30 should expect of
  magnifying number of the resident Internet users,
  as a minimum, on an one-third.

As the whole given problem requires the separate
detailed analysis, as the approximating of Pareto
concerns only to a rich part of population, and
it is more correct to apply to persons with low
ability the logarithmically normal law.
11
13
4. ALGORITHM OF CONTENT ALLOCATION
LONIIS
For the motivation of information resources
decentralization degree the iterative - cyclic
algorithm is offered.
method of one-dimensional search
clusterization algorithm
telecommunication networks planning tools
12
14
The proposed algorithm accounts the following
intuitive reasons
LONIIS

• when number of servers increased, the storage
  cost with other things being equal would be more,
  and the transfer cost would be less
• for different ? ratios the both the boundary
  decision (servers number Ns is equal 1 and
  servers number Ns is equal to node number Nu),
  and the compromise decision Ns ? Nu (? - ratio
  of storage and transfer costs) are possible
• for fixed number of servers the transfer cost
  will be less, if servers will be placed at the
  traffic concentration centers.

13
15
The scheme of search mechanism is the following
LONIIS

• initial value of number Ns(0) of servers is set
• by clusterization algorithm the nodes number
  Ns(0), which have maximum Internet traffic are
  retrieved
• matrix of information gravitation of Internet
  users to servers is constructed
• by redistribution of flows algorithms and
  transfer capacities on the network graph the
  necessary additional values of channel, disk and
  commutative network resources and also their cost
  equivalent are calculated
• according to the selected method of
  one-dimensional search (dichotomizing, "golden
  section" etc.) the new value of servers number
  Ns(1) is determined and return to step 2 is
  carried out.

The algorithms specialization is not mandatory.
The stop condition is dictated by one-dimensional
search method. The problem decision is such a
value of number of servers Ns(i) (i - number of
an iterative steps), at which the minimum of
costs (maximum of profit) is achieved. The
servers geographical positions are defined by
results of cluster analysis. The complexity of
algorithm depends on number of iterative steps,
which one does not exceed 10 for a network with
one hundred clusters.
14
16
As the above formulated problem is related to the
class of optimization problems with a variable
matrix of information gravitation and changed
structure, the following circumstances are taken
into account
LONIIS

• http-connection is initiated by the user agent
  and consists of request, which one at first is
  applied to resources on some initial server
• in general multi-stage request service is
  supposed by nearby server - other local
  servers - external servers
• for matrix dimension decreasing it is expedient
  to represent terminals by clusters.

At cluster analysis for each node i the k nearest
neighbors is set, and the each node i appearance
frequency in list hi is determined. The frequency
is identified with general Internet traffic
incoming from appropriate local and zonal
subnets. After lists regrouping the Ns(1) nodes
are selected with the greatest appearance
frequency. These nodes are taken as servers
locations. Depending on input conditions the
problem is solved on a "node" field or on
"terminal" field.
15
17
5. NUMERICAL EXAMPLE
LONIIS
The problem of the Internet segment of
development interregional Operator is based on
primary and secondary networks is considered at
the following initial conditions

• Internet traffic outgoing from local and zone
  Internet segments produces approximately 5 from
  general telephone traffic and is allocated over
  territory proportionally to densities of the
  telephone subscribers
• secondary telephone network is represented by
  12-tops graph, in which two tops correspond to
  regional class 1 nodes (RC1-X and RC1-Y) (number
  2,11), and remain ones to the automatic toll
  telephone exchanges (fig. 2)
• installation of a local server is permitted on
  any of 12 switches.

16
18
Structure of secondary telephone network
LONIIS
8
6
2
7
3
10
5
9
1
4
11
17
19
It is required to define the character of overall
requirement dependence in an additional network
resources depending of the following variables
LONIIS

• numbers of local servers Ns
• parameter ? is changed in an interval 0 - 1
• part of requests satisfied by local servers which
  are set by k ? r1 / ? r (k 25, 50, 75).

18
20
Some variants are obtained by means of algorithm
given above. Results are represented in the Table
2. Cost interpreting is given in the following
picture. The cost of the project ?? is
represented by the sum of costs on transmission
and storage information.General need in
additional channel resource and the external
requirements to productivity (number of requests
in unit of time) server for different values of
coefficient k.
LONIIS
Table 2
 
19
21
Dependence of general cost ?? from servers
number Ns for different coefficients ? and k
LONIIS
?? (relative units)
 
20
22
The table and figure given above display how the
need for additional channels vary, specific
intensity of requests to a server and project
cost characteristics at servers number
increasing. In particular
LONIIS

• hardware balance is so, that the benefit in
  additional services turns round with necessity of
  additional servers installation
• the greatest resource consumption of channel
  resources is reached at completely decentralized
  structure (number of servers Ns 12) and at
  provision of a maximum of Internet traffic
  closure regional coefficient k. It is explained
  that with addition of each new server will the
  part of traffic ? r1 increase
• on the other hand, with increasing of servers
  number and intensification of their interaction,
  the total number of requests circulating in a
  network increased. However, per server
  request-traffic will be reduced. Thus, the
  minimum number of per server requests is reached
  at Ns 12
• at small specific cost of transmission the
  minimum of general costs is reached on a unique
  server independently of traffic closure
  coefficient
• at small cost of storage the minimum of general
  costs is reached on completely decentralized
  structure Ns 12
• at detail setting in the standard of a complex
  "Planet" the problem of content allocation has a
  multiextreme character, offering a set of the
  identical decisions with different number of
  servers Ns (look at a curve "C", obtained for
  coefficient k 0,5)
• provision of local traffic closure maximum is a
  source of saving of expenditures (compare curves
  "A" and "B", obtained for coefficients k 0,5
  and k 0,75 accordingly)
• saving of costs also is reached by choice of
  appropriate equipment and the network
  architectures (compare curves "B" and "C", having
  identical coefficient k 0,5, but distinguishing
  by coefficient ?, which one for a curve "C"
  less).

 
21
23
6. SUMMARY
LONIIS
The rise of Russian regional Internetization
level sees in rational Internet amplification
in some regions, deployment in others, expansion
of services nomenclature and enumeration of
services, and also strategies improvement of
their providing in third. For today the problem
of content allocation is under steadfast notice
of scientists from many countries. At resemblance
of the given problems with known solutions
proposed approach differs

• reviewing not treelike, but more common
  distributed network architecture
• acceptance of an assumption about probable
  deficiency of network resources
• application of the well-spent topological
  optimization programs for planning structural
  renovation of a network and evaluating of a
  deficiency of its resources
• accounting of QoS-parameters, that is
  automatically provided with algorithms of
  topological optimization and analysis
• an openness and modifiability permitting instead
  of described above algorithms (of one-dimensional
  search, cluster analysis and topological design)
  to apply the apparatus of neuronets and genetic
  algorithms.

 
22
24
LONIIS

• The numerical results of calculations have shown

• in generally, problem of content allocation has
  multiextreme character, offering a set of the
  decisions, identical on the cost, with different
  number of N s servers
• in particular cases at the small specific cost of
  transmission and cost of a storage the boundary
  decisions with number of servers Ns 1 and Ns
  Nu can be obtained
• optimum from the point of minimum general cost of
  implementation the number of servers Ns is
  determined by specific cost ratio of storage and
  transmission information
• general need in channel resource indispensable
  for carry of Internet traffic is the less, the
  more is degree of decentralization structure
  (number of servers)
• at increasing of servers number the average value
  of requests in a time unit on one server is
  reduced
• parameter "the part of requests and other
  interactions satisfied with local servers", has
  an essential value in the view of ordered
  equipment value. By rational organization of
  information structures and telecommunications the
  regional closure traffic coefficient can be
  increased in several times
• analysis of client ability has shown, that the
  given parameters of investments profitability and
  long-term costs on the paths of implantation the
  mass Internet should be provided not only by
  standard economic decisions, but also by
  sponsorship of Internet services at the expense
  of other services and telecommunication services.

 
23
25
REFERENCES
LONIIS
1 A. Koucheryavy. Modern telecommunication
services and prospects for the development of
networks of Russia. Proceedings of Anniversary
Scientific-technical Conference lt the
communications Mans SPb State Telecommunication
University and telecommunication XXI-?? of
century. SPb. 2000. 2 A. Koucheryavy, V.
Nesterenko, A. Paramonov, The strategy of
networks development using new Technology,
International Conference on Communications,
ICC2001 / St. Petersburg, Part 1, pp. 12-17. 3
H.G. Disart, N.D. Jeorganas. New-Clust - an
algorithm for the topological design of two-level
multidrop teleprocessing networks. IEEE Trans. on
Communication, 1977, Vol. Com-25, No10,
pp.1083-1091. 4 L. Kleinro?k, Queueing Systems,
Vol. II Computer application, John Wiley and
Sons, Inc., 1976. 5 A. Koucheryavy,
V.Lokhmotko, A.Paramonov. Network simulation
system. ISTN ' 2000 Proceedings 2000
International Seminar on Teletraffic and Network.
Teletraffic Issues and Technology on Broadband
Telecommunication. Nov. 15-17, 2000. Haangzhou,
China. 6 H. A. Taha, Operational Research. An
Introduction, MacMillan Publishing Co., Inc.,
N.Y., 1982. 7 I. Cidon, S. Kutten, R. Soffer.
Optimal allocation of electronic content.
Computer Networks, Vol. 40. Iss.2. 2002, pp.
205-218. 8 E. Balas, W. Niehaus. Optimized
crossover-based genetic algorithms for the
maximum  cardinality and maximum weight clique
problems. J. Heuristics. v4. (1998), N4, pp
107-122.
 
24
26
LONIIS
Thanks for your attention