www.seipub.

org/ijc

International Journal on Communications (IJC) Volume 3, 2014

A Novel Mathematical Model to Calculate

Steady-state Throughput of the Bulk Transfer
TCP Flows
*1M M Hassani, 2Mohammad Enayattabar, 3Meysam Mohammadi
Iran Azad University Science and Research Ayatollah Amoli Branch, Amol, Iran
*hassanimm@yahoo.com

Received Nov 16, 2013; Accepted Jan 16, 2014; Published Jul 9, 2014
2014 Science and Engineering Publishing Company
Abstract
Wlan networks have a great influence on human
everyday life. You can see its usage in airport, wireless
transmission like mobile, to prevent fire in jungle,
communications, est. but what is the relation between
math and network? Network is a simulation model of
graph and the throughput of its many parameters is
measured by analytical models .one of this math
application in network fields is markov model. We
evaluate extensively through analysis, simulation, and
experimentation the interaction between the TCP
flows. The main goal of this paper is calculating the
throughput ratio between upstream and downstream
TCP flows. At last, we demonstrate a mathematical
model to analyze the behavior of up/down flows and
we prove that it can be used in real models.
Keywords
Short Term Flows; TCP; WLAN
Introduction
Short Flows are TCP flows which will last less than one
round-trip time (RTT) without the overhead of TCP
congestion control. With feedback based TCP
congestion control techniques, the slow start period
lasts a few RTTs and therefore results in tremendous
underutilization of available (and continuously
improving) network bandwidth. In the short span of
four years since becoming a standard, IEEE 802.11b
Wireless Local Area Networks (WLANs) look set to
revolutionize the way people perceive access to
networks. By becoming a ubiquitous facility, available
both in public areas such as airports, cafes, museums,
workplaces, and in private homes, it is conceivable that
people will soon expect that the same (or a

38

comparable) QoS they receive on wired networks are

also available to wireless networks. To wit, one could
argue that sayingwired networks andQoS in the
same breath has often been a debatable issue. The
provision of QoS guarantees has often been hindered
by several factors including, but not limited to, the
inherently best-effort service model of the IP protocol,
the uncertainty in characterizing the traffic generation
process, the unavailability of QoS routing algorithms
and protocols that could be deployed on a global basis,
and the poor scalability of several attempts at defining
an articulate QoS paradigm. Among the several
solutions proposed in the literature in the past years,
the most successful one (both in terms of consensus
and of actual deployment) has to be the DiffServ
architecture [1]. Diff- Serv received the endorsement of
IETF and routers today offering a wide range of
support for DiffServ capabilities.On top of these
observations, wireless networks offer a whole set of
new challenges to tackle. Traditionally, obstacles to
QoS provisioning in wired LANs are: their distributed
nature, the burstiness of traffic,the unpredictability of
the number of users, the random access protocol.
Beside these, WLANs using the 802.11b technology
further add some of their own, such as poor channel
quality depending on relative position of Wireless
Stations (WSs), interference from hidden terminals,
and the anomaly observed in [2], due to the different
speeds at which WSs transmit. This lack of
determinism has several drawbacks. Among them, the
guarantees, albeit of a statistical nature, that
architectures like DiffServ provides cannot be
extended to users of wireless communities, because
even the most privileged traffic flows (in wired parts
of the network) can be unpredictably delayed or
discarded by temporary channel failures. Additional

International Journal on Communications (IJC) Volume 3, 2014

mechanisms are therefore necessary if QoS guarantees

are to be provided. If an extension of DiffServ
capabilities is to be engineered for wired-cum-wireless
networks, two issues are prominent: differentiation of
traffic classes (and its support in network elements)
and fair access to network resources for all flows
within the same traffic class.
In this paper we are going to describe casseti algorithm
to evaluate throughput of short term TCP flows. First
we describe this method and explain its algorithm. In
the third section we demonstrate our
proposed
model, and finaly , we compare our model against
casseti algorithm by many throughput parameters.
Casseti Alghoritm
The goal of our scheduling algorithm is twofold:
firstly, to improve the fairness among wireless stations
that may experience location-dependent channel
capacity and errors; and, secondarily, to provide shortterm fairness to
TCP traffic in order to enhance the performance of
shortlived flows.
To achieve these results, we proceed as follows. At the
AP LLC layer, we introduce a separate queue for each
WS associated with the AP, while only one queue is
implemented at the WS LLC layer. A channel
conditionestimator is associated with each queue, and
transmission is allowed only for those queues whose
channel is estimated to be good, i.e., such that 11 Mbps
speed can be attained. Queues attempting to access the
wireless medium with a bad channel will be allowed to
transmit again when their channel becomes good.
Upon switching to good channel state, a queue that has
just experienced a bad channel is rewarded with the
possibility to sendto the MAC layer a maximum
number of back-to-back frames (hereinafter referred to
as TXburst), proportional to its forced silence period.
The MAC layer will then use the EDCF bursting
capability2 of the current 802.11e draft [18], to avoid
contentions within the burst3.
The building blocks of our algorithm are the following:
BB channel state estimation B queue selection and
service TXburst length setting.
The algorithm is implemented both at the AP and at
WSs. Note that, since only one queue is used at each
WS, the queue selection is not needed at WSs. Below,
we briefly describe these building blocks for the most
general case, i.e., for the AP.A. Channel State

www.seipub.org/ijc

Estimation. As mentioned above, the AP estimates the

channel conditions for each contending WS. Thus, a
flag is associated with each LLC queue at the AP,
indicating the corresponding channel state. The flag
can take three values: GOOD, BAD or PROBE.
GOOD: The AP sets the flag to GOOD whenever it
receives, from the corresponding WS: (i) a MAC-layer
acknowledgment in response to a data frame, (ii) a
CTS frame in response to an RTS frame, or (iii) an
error-free data frame or RTS.
BAD: The AP sets the flag to BAD after a transmission
failure. To tell if the cause of a transmission failure is
due to collisions or channel errors, we resort to a
careful selection of the values for the Short Retry Limit
(SRL) and the Long Retry Limit (LRL). These
parameters control the number of transmission
attempts without receiving an acknowledgement; the
number of attempts is tracked by a Long Retry
Counter (LRC) and a Short Retry Counter (SRC). In all
likelihood, the LRC is incremented when the
transmission of a frame longer than the RTS threshold
fails due to channel errors [13].
Instead, the SRC is incremented both because of
channel errors and because of collisions over the
shared medium. When LRC (SRC) reaches the LRL
(SRL) value, the MAC layer abandons the transmission
of the frame and it signals the failure to the LLC layer.
While LRC increments allow a detection of channel
errors, the interpretation of SRC increments is
ambiguous. We then apply an empirical method: we
perform some simulations with a large number of
contending stations 2Setting the TxOpLimit to the size
of the burst received by the LLC layer.
3The EDCF bursting capability provides contentions
only for the first frame of the burst, while each
subsequent frame in the same burst can be transmitted
after the ACK of the previous frame without
contention. If an ACK is not received, the burst
transmission is aborted.
Fig. 1 Queue service (top) and queue selection (bottom)
flow charts. The communication and data management
between MAC layer and LLC layer occur following the
standard OSI paradigm of request/indication. The LLC
layer sends a transmission request to the MAC layer
along with the frames, belonging to the same TXburst,
which are to be transmitted. These frames also remain
in the LLClayer buffer in case retransmission is
required. The MAC layer then attempts the
transmission following the EDCF 802.11e burst

39

www.seipub.org/ijc

transmission technique. If all frames in the TXburst are

successfully transmitted (i.e., the last MAC-layer ACK
of the burst is received), the MAC layer returns a
frame success indication to the LLC layer.
If a transmission error occurs, the MAC layer discards
the remaining frames in the TXburst; these frames are
however stored at the LLC layer. The MAC layer then
returns a frame failure indication to the LLC layer,
together with the number of successfully transmitted 6
frames in the current TXburst, which can be discarded
at the LLC layer. When the MAC layer supplies a
frame success indication to the LLC, the following
cases may occur: (i) if the flag of the queue that is being
served has a GOOD value, the next queue is polled; (ii)
if the queue flag has a PROBE value, the flag is set to
GOOD and the queue is enabled to start a TXburst
whose length is determined according to the
procedure outlined in Section IV-C. If the queue is
empty, the next queue is polled. When the MAC layer
notifies a failure, the flag is set to BAD and the next
queue is polled. Polling a queue involves checking
both whether it is empty and the value of the
associated flag. In case of not empty queue, (i) if the
flag is BAD , the PTIMER is inspected: if still running,
the next queue is polled; otherwise, the flag is set to
PROBE, and the queue is served. Likewise, (ii) if the
flag is PROBE, the queue is served. Finally, (iii) if the
flag is GOOD, the TXburst length is computed and the
queue is served. We also remark that each frame has
an internal timestamp associated with it: frames that
have been waiting in the queue for a time greater than
a STALE TIMEOUT value are discarded. We
introduced this timeout to avoid an excessive latency
during long channel error periods. We found by
simulation that setting STALE TIMEOUT to 1 s allows
the average end-to-end delay to remain below
reasonable values.

International Journal on Communications (IJC) Volume 3, 2014

Connection Division Scheme

In this section we run the simulations using the same
simulation methodology mentioned in section 4.1.The
difference here is that we have two different
connections; one from fixed host to the base station
and the other from base station to the mobile host. We
used two, three and six wireless networks in these
simulations. The dropping probabilities for these
wireless networks are 1%, 2%, 6%, 4%, 6% and 2%
respectively.
Figures 10, 11, and 12, show that the average
completion time increases as the session size (in
segment) increases.

Queue Selection and Service

The process of queue selection and service at the LLC
layer relies on indications from the MAC layer
regarding the success or the failure of a frame
transmission on the downlink. At the LLC layer,
queues are served in a round-robin fashion. The queue
selection algorithm is sketched in Figure 4, where two
macro-blocks are outlined. The top one describes the
actions taken when frames are sent to the MAC layer
and the transmission outcome is notified to the LLC
layer; the bottom one describes the queue selection.

40

FIGURE 1 QUEUE SERVICE ND QUEUE SELECTION FLOW

CHART

Conclusions
In this paper, we focused on a very significant problem
of wireless LAN called short term tcp flows.
Researcher introduced many methods to increase the
throughput of these flows. As it is observable, many
appliv\cation layer protocols (like http) have a short
time interval for communicating. This short term flows

International Journal on Communications (IJC) Volume 3, 2014

may be located after a long term flows and their

proiority may be low because of this time. We
proposed a recursive and analytical model for shortlived TCP flows and we focused on heterogeneous
wireless networks that have different packet dropping
probability. We proposed this model for Connection
Division scheme. The Local Retransmission scheme
was not modeled because of the complexity of
equations. The average completion time in the
proposed wireless TCP model for heterogeneous
wireless networks is calculated to determine the TCP
performance using the average round trip time,
retransmission timeout, flow size and the segment
dropping probability.
The results obtained show that the simulation and the
model values are relatively close to each other with the
percentage of difference for the average completion
time (in Sec) being less than 3% and 6% for the End-toEnd scheme and the Connection Division scheme,
respectively.

www.seipub.org/ijc

proposed model for the two wireless schemes reflected

TCP behavior well, specifically the session completion
time at the initial stage (i.e. slow start).
REFERENCES

H. Wang, C. Williamson; A New Scheme for TCP

Congestion

Control:

Smooth

Start

and

Dynamic

Recovery; Proceedings 6th International Symposium on

Modeling, Analysis, and Simulation of Computer and
Telecommunication Systems pp 69-76; 1998
J.R. Chen, Y.C. Chen; Pseudo-Rate TCP: A Congestion
Avoidance Scheme with Nearly Optimized Fairness and
Throughput; Computer Communications pp 1493-1501;
1999
K. Thompson, G.J. Miller, R. Wilder; Wide Area Internet
Traffic Patterns and Characteristics; IEEE Network Vol.
11 Issue 6 pp 10-23; 1997
L.S. Brakmo, L.L. Peterson; TCP Vegas: End-to- End

Finally we are going to compare our method with

casseti model.

Congestion Avoidance on a Global Internet; IEEE

Journal on Selected Areas inCommunications Vol. 13
Issue 8 pp 1465 1480; 1995
M. Allman; TCP Congestion Control; RFC 2581;
M. Bottigliengo, C. Casetti, C.-F. Chiasserini, M. Meo short
term fairness for TCP flows in 802.11bWLANS ; 1995
M.T. Lucas, D.E. Wrege, B.J. Dempsey, A.C. Weaver;
Statistical Characterization of Wide Area IP Traffic;
Proceedings 6th International Conference on Computer

FIGURE 2 TOTAL OF THROUGHPUT WHEN MOBILE STATIONS

INCREASE FROM 1 TO 40

Comparison result shows that our proposed model has

smoothy curve against casetti model and when the
number of mobile stations increases, the total
throughput of short term flow has more value than
that of casetti model and it means that: The
Transmission Control Protocol (TCP) is widely used in
the Internet world and several analytical models were
proposed to improve the performance of TCP in terms
of throughput. However, these models cannot be used
for short-lived TCP flow which is considered as the
most important characteristic in the Internet. We used
the NS-2 simulator to verify the accuracy of our model
and we compared the simulated result for
heterogeneous wireless topology with the calculated
values that would be obtained from our proposed
model.
Based on the obtained results, we found that the

Communications and Networks pp 442- 447; 1997

R. Chang, L. Huynh, J Gray; Adaptive Rate-Based
Congestion Control Versus TCP-SS; Proceedings 1993
International Conference Network Protocols pp 186-197;
1993
R. Fox; TCP Big Window and Nak Options; RFC 1106; 1989
R. Wade, M. Kara, P.M. Dew; Proposed Modifications to
TCP Congestion Control for High Bandwidth and Local
Area

Networks;

6th

IEE

Conference

on

Telecommunications pp 151-155; 1998

V. Jacobson; Congestion Avoidance and Control; ACM
SIGCOMM 88 pp 273-288; 1988
V.

Jacobson;

Modified

AvoidanceAlgorithm;

TCP

mailing-list,

Congestion

end2end-interest;

1990
W. Stevens; TCP Slow Start, Congestion Avoidance, Fast
Retransmit, and Fast Recovery Algorithms; RFC 2001; 1997

41