|
FEATURED GOLDSIM APPLICATION
Modeling Pricing Mechanisms in Packet-Switched Communication
Networks Using GoldSim
Tomaž Turk
Faculty of Economics, University of Ljubljana, Slovenia
Introduction
In recent years, several pricing models which address the
problem of Internet congestion have been studied. Today Internet
users are charged mostly on a flat-fee basis, regardless of
the network load they introduce to the network. Different
Quality of Service (QoS) models are being proposed in the
literature, such as the price-controlled best-effort model,
which introduces the general idea of usage-sensitive or variable
pricing. Other possible models include per-packet or per-volume
flat rate pricing, and flat rate pricing dependent on the
QoS class, to name just a few. In practically all proposed
models, a price is established for each network connection
(sometimes also for data buffers on network nodes, i.e., routers).
One of the issues for the proposed pricing mechanisms is
the calculation of the total cost of data transfer, which
is basically the sum of the costs incurred along the network
path, or in other words, the summed costs for data transfer
over each node-to-node connection. One possible solution would
be to include the information about the highest price the
user is prepared to pay per transferred MB ("bid price")
over the entire path in his data stream. For each step along
the route, the network node which controls data entry onto
the connection would subtract the connection price from the
price indicated in the data stream. This would be repeated
along each path the data stream is passing. It is possible
that the highest price is too small to cover the entire route,
so the proposition assumes at least two QoS classes - paid
and unpaid traffic. The user pays according to the actual
connection price, not according to his bid.
The question is, however, whether the proposed schema will
work in practice. For this, the stability of the system should
be examined. If the basic model gives acceptable estimates,
the model can be used to test different pricing policies.
The Simulation Model
Communication networks are being explored mostly using discrete
event simulations and queuing systems. In our model, the traffic
in the communication network is represented as the flow of
data between network nodes. This gives us the possibility
to use a different simulation approach, namely the systems
dynamics simulation methodology. Since GoldSim enables the
simulation of discrete events, we could model discrete influences
too (e.g. the change of routing rules, broken connections
etc.). An important aspect of our model is the bid price,
which is embedded into the data flow. In the model this is
represented as a parallel information flow, and is referred
to as the value flow.
Basically, each network node (router) could be represented
as a data buffer (a Reservoir in GoldSim), with the connection
to another node. We chose a different approach, where flows
between network nodes are represented in matrix notation.
For this to work, we introduced the following terms:
Source (data traffic generator; e.g., a user);
Connection (physical connection between two network
nodes; e.g., optical fibre);
Link (virtual connection between two network nodes
or between a source and a node; a part of the network path
the traffic from a particular source is passing);
Node (data buffer for the connections).
Each source introduces some traffic flow onto the network,
together with its value. Both streams (data and value) are
flowing through the network along the path, which is represented
as a series of links. Links share common connections. Each
connection has its capacity (bandwidth) and price.
The network topology is given as a set of matrices:
sources to links matrix;
links to links matrix;
links to connections matrix; and
links to nodes matrix
The first and the second matrices define the basic traffic
paths. The third matrix represents connection sharing among
links. The last matrix represents the way links share network
nodes. (This could have been represented in another way, with
the "connections to nodes" matrix).
For instance, the matrix Link2Link in Figure 1 shows that
the first link is connected to the second, the second is connected
to the third, and the third is connected to the fourth link.
Figure 1: The network topology is represented in matrix notation.
The model elements are grouped into four modules (GoldSim
Containers), where the most complex one is the Container that
represents the communication network. Other Containers represent
such parts as source behavior and parameters of the model
(such as the above mentioned topology). The structure of the
network model can be seen in Figure 2.
Figure 2: The main structure of the network model.
The top part of the model represents the data flows, while
the bottom part models the value flows. (The traffic introduced
into the network by the sources is modeled as being stochastic.)
The middle part of Figure 2 includes the calculation of output
flows (both data and value flows), together with the traffic
measurements and price adjustments in the case of variable
pricing. Figure 3 shows the details of the network node model,
which includes such estimations and calculations as buffer
capacities, dropped packets, the delay on network node, etc.
The model tracks data and value flows for each source, network
node, and connection in time.
Figure 3: The details of network node model.
Figures 4, 5, 6 and 7 give the results from running one realization,
where a single connection (and link) is shared among 20 sources.
Each source introduces 2 Mbps of unpaid traffic on average.
The connection capacity is 200 Mbps, so it is at its limits
(the network is in congestion state). The first user (source)
decides to pay for his traffic, and the price he is willing
to pay is at most 80 $ per MB. We can see from Figure 4 that
the connection price is sometimes above his bid. (The connection
price is variable; it is recalculated each second, which introduces
relatively quick and drastic changes to its value.) Figure
5 represents the data stream from the first user for which
he is prepared to pay. We can see that in time intervals where
the connection price is higher than his bid, the data stream
changes to another (unpaid) QoS class.
Figure 4: The connection price.
Figure 5: Paid network traffic for the first user.
Figure 6 shows the data traffic overflow rate for the first
user. Data traffic overflow rate represents dropped packets,
and is dependent upon the traffic state (congestion), the
buffer capacity on network nodes, and QoS class. When the
user pays for his traffic, the dropping of packets doesn't
occur (paid traffic has a higher priority than unpaid one).
Figure 7 shows data traffic overflow rate for his fellow user,
who is not prepared to pay for his traffic, and is otherwise
behaving in similar fashion to the first user.
Figure 6: Traffic overflow rate (dropped packets) of the first
user (mostly paid traffic).
Figure 7: Traffic overflow rate (dropped packets) of the second
user (unpaid traffic).
Conclusion
From the above we can see that the communication network
can be successfully modeled with the system dynamics approach.
Our model can be used for other purposes too, since the pricing
mechanism, i.e. the lower part of the structure on Figure
2, can be excluded from the model. Nevertheless, the above
model is quite complex, particularly the calculation of data
flows from network nodes to connections for each source. Other
relatively complex parts of the model include value flow calculations,
buffer capacities distribution between sources, and price
determination when variable pricing is tested.
About the Author
Tomaž Turk is an economist and has a PhD in information
sciences. He is an assistant professor and researcher at the
Universtity of Ljubljana, Faculty of Economics. He teaches
Development of Information Systems, Economics of Information
Technology, Economics of telecommunications, and
Business Simulations. Currently his research work includes
themes from communication networks management, internet society
issues and economics of information systems.
|
