International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 179

ISSN 2229-5518

L4-DMF: An Enhanced Self Adaptive Congestion

Control Mechanism in Layer-4, a Simulation


1Azubuogu A.C.O, 2Ochi Josiah .C, 3Ugwoke F.N, 4Okafor K.C, 5Udeze C.C.


Electronic and Computer Engineering, Nnamdi Azikiwe University Awka,


Computer Science, Michael Okpara University Umuahia, Nigeria,


Electronics Development Institute, Awka, Nigeria,



edujosh09@ yah




udezechidiebele@, ,

Abstract— As part of an ongoing research, this work presents an enhanced self adaptive congestion control mechanism in layer-4, called Dynamic Modulation Feedback (DMF) which can fit into the network condition dynamically according to the parameters given by upper layer and the network layers These layers comprise of buffer windows, data rates, Average Queuing Length (AVQL), channel conditions or Packet Error Rate-PER. In this work, an optimization model for throughput maximization in Flow Aware-W ireless Local Area Network (FA- WLAN) was developed. A simulation was carried out with OPNET IT Guru to compare the congestion control mechanism of TCP- Tahoe, Reno, New Reno, Vegas and Sack (TCP-TRONVS) with a proposed layer-4 TCP-DMF. Performance metrics such as Latency, Throughput, Buffer Utilization, and Packet Loss Ratio, was used in the analysis while the measurement results from OPNET Modeler 9.1 was analysed. Consequently, this work showed that with the proposed layer-4 DMF congestion control mechanism, FA-W LANs will scale gracefully in this new era of High Performance Computing (HPC).

Keywords — Channel, Congestion, Layer-4 DMF, Optimization, Packet Loss Ratio, Performance, TCP-TRONVS


ONGESTION management in wired and wireless envi- ronments has continued to attract attention in the net- working market segments. The Internet topology is changing fast, and more often than not includes a Wireless last Hop. The number of internet services increases rapidly every year like TCP Hypertext Transmission Protocol (HTTP), Do- main Name Service (DNS), Hypertext Transmission Protocol Secured (HTTPS), etc. A user has a wide range of possibilities, not only in choosing the applications, but also in selecting the end user devices or the connection method [1]. Today, packets may get lost on wireless links, for instance due to radio inter-
This is a dramatic change over the assumption that was made
in wired networks, where most (if not all) of packet losses
were due to network congestion or buffer overflow [2]. While
the congestion control mechanisms of TCP-TRONVS offers a
good stability for inelastic applications and traffic flows,
adapting WLAN in high performance computing requires an
extensive study of its congestion management schemes while
proposing a possible enhancements.
In data networking andqueuing theory, network conges-
tion occurs when a link or node is carrying so much data that
its quality of service deteriorates. Typical effects include queu- ing delay, packet loss or the blocking of new connections. A consequence of these latter two is that incremental increases in offered load lead either only to small increase in net-
workthroughput, or to an actual reduction in network throughput [3].
Network protocols which use aggressive retransmissions to compensate for packet loss tend to keep systems in a state of network congestion even after the initial load has been re- duced to a level which would not normally have induced
network congestion. Thus, networks using these protocols can exhibit two stable states under the same level of load. The sta- ble state with low throughput is known as congestion collapse [3] while we refer the stable state with high throughput as congestion upgrade. This work defines WLAN congestion collapse as a condition which a packet switched Access Point (AP) halts traffic flow to congestion. Congestion collapse gen- erally occurs at choke points in the network, where the total incoming traffic to a node exceeds the outgoing bandwidth. In a heterogeneous link, connection points between a local area network and a wide area network are the most likely the con- gested points. When a network is in such a condition, it settles (under overload) into a saturation stable state where TCP traf- fic demand is high but little useful throughput is available, and there are high levels of packet delay and loss (caused by routers discarding packets because their output queues are too full) and general quality of service is extremely poor. Generally, Transmission Control Protocol (TCP) which utilizes a network congestion avoidance algorithm that includes vari- ous aspects of additive increase/multiplicative de- crease (AIMD) schemes, with other schemes such as slow-start in order to achieve congestion avoidance. The TCP congestion avoidance algorithm is the primary basis for congestion con- trol in the Internet [4], [5]. To avoidcongestion collapse, TCP uses a multi-faceted congestion control strategy to handle lay- er-4 issues which is not ideal. However, TCP-TRONVS is very sensitive to packet losses and requires further improvements to better adapt to the wireless environments. A good conges- tion control scheme should be able to dynamically handle packet arrivals at the network edges, hence allowing fair shar- ing of resource among all the users leading to a significant bandwidth optimization.

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 180

ISSN 2229-5518

1.1. Our Contribution

Congestion control is the problem of managing network traffic or a network state where the total demand for resources such as bandwidth among the competing users exceeds the availa- ble capacity. This work views it as a layer-4 problem in the OSI reference model and it is a core infrastructural problem stemming from the packet switched and statistically multi- plexed nature of the Internet and has an impact on the Internet stability and manageability for realistic loads. Hence, this work proposed an improved congestion management scheme called TCP layer-4 Dynamic Modulation Feed (L4-DMF) which will address both packet losses even at busty traffic and congestion in a FA-WLAN.


The Flow-Aware Networking concept, first proposed in 2004 [6] is relatively new. The main goal of FAN is to achieve effi- cient packet transmission with the minimal knowledge of the network. In FAN, the traffic is sent as flows (streaming or elas- tic). The first type of flows is usually used by real-time appli- cations while the second one carries best effort traffic. Two scheduling algorithms were proposed for FAN, PFQ (Priority Fair Queuing) [7] and PDRR (Priority Deficit Round Robin) [8]. Another proposal for realizing the FAN concept, called AFAN (Approximate FAN), was described in details in [9]. These FAN versions yield similar results. That is why the sim- ulation analysis presented in this paper is provided only for the PFQ algorithm. FAN is a scalable solution. The work [10] opines that the complexity of the queuing algorithms does not increase with the link capacity. Moreover, fair queuing is fea- sible, as long as the link load is not allowed to attain saturation levels (it is ensured by the admission control).
According to [11], four congestion control mechanisms have been proposed for Flow aware Networks, FAN i.e., EFM (En- hanced Flushing Mechanism), RAEF (Remove Active Elastic Flows), RBAEF (Remove and Block Active Elastic Flows), and RPAEF (Remove and Prioritize in access Active Elastic Flows). The work in [12] argues that TCP congestion avoidance algo- rithm is the primary basis for congestion control in the Inter- net [12]. The work in [13] used simulations to explore the ben- efits of adding selective acknowledgments (SACK) and selec- tive repeat to TCP. It compared Tahoe and Reno TCP, the two most common reference implementations for TCP, with two modified versions of Reno TCP. The first version is New-Reno TCP, a modified version of TCP without SACK that avoids some of Reno TCP's performance problems when multiple packets are dropped from a window of data. The second ver- sion is SACK TCP, a conservative extension of Reno TCP mod- ified to use the SACK option being proposed in the Internet Engineering Task Force (IETF). The work then described the congestion control algorithms in a simulated implementation of SACK TCP while showing that selective acknowledgments are not required to solve Reno TCP's performance problems when multiple packets are dropped. Again the absence of
selective acknowledgments does impose limits to TCP's ulti- mate performance. The work concluded that without selective acknowledgments, TCP implementations will only be con- strained to either retransmit at most one dropped packet per round-trip time, or to retransmit packets that might have al- ready been successfully delivered. The congestion control and packet retransmission algorithms in Tahoe, Reno, New-Reno, and SACK TCP will be presented in the next section.
The author in [14] presented a study on congestion control and scheduling in communication networks. In contrast to stand- ard protocol design where there is minimal communication between the scheduling and the congestion control, the paper argued that there are a number of benefits to jointly optimiz- ing these algorithms, especially in wireless networks. The first, the benefit of the coordination is that effective buffer sizing, is achieved even when channel rates are variable in the WLAN. The second benefit from [14] is that we can prevent conflict situations where the congestion control and the scheduler both try to assign bandwidth to the flows. The third benefit is that coordination allows us to prove theoretical utility maximiza- tion results that are not affected by possible oscillations.
In analysis done by Kelly, Maulloo and Tan in [15] it was shown that many variants of TCP can be viewed as approxi- mations of primal-dual algorithms that solve an underlying optimization problem. However, in most cases as studied in literature, this analysis abstracts away some of the details of the scheduling problem. From this research, it was observed that congestion control mechanism implemented through Queue Management algorithms, is considered a key factor to solve this optimization problem above while keeping TCP/IP networks efficient and reliable from the user's viewpoint.
Many works in literature have investigated methods to esti-
mate the key wireless link parameters in the context of conges-
tion metrics, for example, link bandwidth estimation [16], link
buffer size estimation such as max-min, loss-pair and sum-of
delays [17], [18], and queue length estimation [19]. For this
reason, the utilization of fuzzy logic (FL) has shown to be use-
ful in designing new active queue management (AQM) meth-
ods [20] that can be used to alleviate congestions in wired and
wireless networks. For example, Mallapur et al. [21] proposed
a buffer manager located at the base station using a fuzzy con-
troller for packet dropping in wireless cellular networks. The
controller uses three fuzzy parameters, namely application priority, queue length and packet size.
Many researches have been devoted on the queue size man- agement algorithms such as Random Early Detection (RED) and Weighted RED, PID controller [22], and NLRED to reduce the likelihood of global synchronization, as well as keeping queue sizes down in the face of heavy load and bursty traffic. Other related AQM schemes such as GRED, WRED, and ARED are summarized in [23]. The Random Early Detection (RED) mechanism can solve the Drop Tail's (DT) deficiencies well. It uses randomization to ensure that all connections en- counter the same loss rate. By dropping packets before the router's buffers are completely exhausted, the RED mecha- nism try to prevent congestion. In [24], another approach known as Random Exponential Marking (REM) was also de- veloped to measure the average queue size instead of conges- tion measure. It was able to achieve high link utilization, neg-

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 181

ISSN 2229-5518

ligible packet loss and short queuing delay in a simple scalar manner. TCP assumes that losses of packets are derived from a buffer overflow at a Base Station (BS) router. It invokes the congestion control to reduce traffic unnecessarily when the packet loss occurs due to transmission errors on a wireless link. Thus one approach to mitigate this problem and distin- guish packet loss due to congestion and others packet errors due to channel variations, is to employ TCP proxy called Per- formance Enhancing Proxy (PEP) [25]. However, the major problem of such Indirect-TCP (e.g. Snoop protocol) and M- TCP is a hand-off latency due to mobility of (mobile host) MH [26], [27].
The observed limitations of the various congestion manage- ment schemes studied in sample literatures is outlined as fol- lows:
1. The quantitative evaluation of the performance of TCP over 802.11 WLANs shows that TCP congestion control schemes cannot handle packet losses effective- ly in a FA-WLAN for high performance computing.
2. There is no discussion on real physical parameters of
the realistic loads and feasible network parameters for
congestion control in FA-WLAN.
3. The congestion management schemes owing to the
structure of their algorithms introduce latency, there-
by affecting the traffic throughput.
4. There is no dynamic provisioning on the network
edge devices such as APs to manage queues regard-
less of the data rates.
5. There is unfairness in throughput delivery as the
throughput for nodes depends on the number of
flows that it has (TCP oriented).
6. Active Queue Management schemes focus more on
packet drops on exceeding the equilibrium threshold
rather than packet fragmentation or implicit feedback
(such DMF)
Consequently, a FA-WLAN model for high performance com-
puting must demonstrate its capability in addressing these
limitations as observed in context.


3.1. FA-WLAN Congestion Management

As discussed in our earlier work, the observed limitations of TCP-TRONVS in IEEE 802.11b Wireless Networks gave rise to the proposed TCP layer 4-DMF. A fuzzy logic implementation in a rule editor of MATLAB 2009b which is referred to as Fuzzy-Logic Adaptive Queuing controller (FLAQ) inherits from the classical Random Early Detection (RED) algorithm. TCP layer 4-DMF has an implicit dynamic modulation feed- back. As a variant of TCP FLAQS, this was adapted in FA- WLAN for congestion analysis. As shown in Fig. 4.1, the fuzzy DMF controller predicts dynamically the packet dropping rate (joint scheduling and congestion control) and the correspond- ing average queue length. It relies on the average queue length at the base Access Point router (AP) and the packet loss rate caused by the channel variations in mobile environment; as-

suming there is no buffer overflow due to the congestion. Us- ing this model, a heuristic TCP performance can be estimated over a time-varying channel under different conditions of us- er’s mobility.
Fig.1: TCP layer-4DMF mamdani fuzzy inference engine
When the arriving packets cannot be accommodated due to lack of network resources (bandwidth, buffer size, etc), this indicates congestion occurring at router buffers of networks. More specifically, a poor network performance due to conges- tion can be offered in terms of high dropping and queuing delay for packets, low throughput and not maintaining the average queue length which may not prevent the router buff- ers from building up, then dropping packets.
However, congestion control mechanism implemented through intelligent Queue Management algorithms, is as- sumed to be the key factor to solve this problem keeping TCP/IP networks efficient and reliable from the user's view- point.
The result of the proposed layer 4-DMF showed a significant improvement in TCP QoS (throughput, etc) performance con- sidering the user’s mobility and realistic traffic loads in the FA-WLAN.


4.1. Management System Representation/Planning

Given a set of AP equipment and a set of available channels, the objective function is to design an efficient network plan for a FA-WLAN such that maximum wireless coverage of a speci- fied area is achieved and that congestion in the wireless topol- ogy (BSS) is efficiently managed. In this context, the efficiency of a network plan Ef is defined by a channel overlap measure and the net throughput expected over the wireless service area for a single user in the FA-WLAN.
Let A be the set of candidate AP locations, and let M (0 < M≤│A│) denote the maximum number of APs to be deployed and positioned. The specific locations are determined in ad- vance with respect to factors such as potential installation costs, accessibility, physical security, radio propagation as- pects, health safety, psychological factor, etc. Fig. 4.2 shows

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 182

ISSN 2229-5518

the AP LTP used in this research.

The active service area is represented by a grid of Localized Test points (LTPs) with a given resolution. Let J denote the set of LTPs to be covered. The received signal strength at LTP j depends on the transmission power level TPL (APi ) at the serving APi and attenuation Atn between the AP and LTP j. The received signal strength be computed as:

Rss j (rec):Atn .TPL (APj )

The transmission power TPj (APj ) is assumed to be fixed and ascertained in advance. The attenuation Atn is given in a path-
loss prediction grid in channel model shown in Fig.1

4.2. FA-WLAN Design Assessment

4.2.1. Coverage

A LTP j (j€J) is covered if there is at least one installed APj (j€A) such that the received signal from j satisfies RSSj (rec) Ø(r), where Ø(r) is the receive sensitivity threshold (RST). This parameter Ø(r) defines the minimal signal strength required for receiving transmissions at the lowest possible data rate. This threshold can be adjusted as it is an adjustable configuration parameter with its typical value found in the APs documenta-
The AP placement determines the maximum data rate that a user can expect. With growing distance to the closet AP, the net data rate decreases as shown in Fig. 4.2. The max possible data rate is only achieved if a user does not have to contend with the medium. Under this assumption, maximizing the average data rate taken over all TPs allows the expected user’s throughput. The goal in this context is to maximize the throughput that a user can expect. This leads to a problem formulation in integer programming model viz:

Given a typical facility location model, how can the problem of max- imizing the expected throughput with M most installed AP be real- ized?

From the formulation above, in a typical facility location mod-
el in FA-WLAN, let us use two classes of binary variables vz:

variables 𝑧𝑎 𝜖{0,1} for all potential locations in a, and varia- bles𝑧𝑎𝑗 𝜖{0,1}, for all pairs of locations and TPs. Here Za =1

suggests that an AP is installed at location a. Again, Xaj = 1
means that TPj is associated to APa. It is worthy to note that variable Xaj can be set to zero if the received signal strength in TPj from APa is below received sensitivity threshold Ø(r).
Our objective function in maximizing throughput in a con- gested network is given by:


tion manual. In this context, a network with very extensive


.𝑎. 𝑗∅(𝐴𝑡𝑛 ) 𝑥𝑎𝑗 (1)
coverage is desired in the FA-WLAN.

4.2.2. Throughput

To ascertain our throughput, OPNET IT Guru is used to find a fitting function that represents the throughput experienced by a user in a FA-WLAN environment. The throughput depends


Subject to:

𝑥𝑎𝑗 ≤ 𝑧𝑎 (2)
𝑎 𝑥𝑎𝑗 ≤ 1 (3)
𝑎 𝑧𝑎 ≤ 𝑀 (4)
on the strength of the strongest signal received by any AP, hence it is assumed that the transmit power to be fixed on the

Z𝜖{0,1}P , x 𝜖 {0,1}



attenuation value. Given that a client AP is denoted by a, then the net throughput of the user j is denoted by Ø (TPLj ).

The throughput in some area is a strong indication that a bet- ter AP placement is needed. Again, our new objective function is to maximize the total throughput over all LTPs by choosing an appropriate subset of candidate AP locations.

4.3. Optimization Models

The goal in our FA-WLAN planning and design is to maxim- ize the throughput that a user can expect regardless of the traf- fic type. Two aspects are considered viz:
a. Data rate (contention Quality at the physical lay- er)
b. Contention for the medium with other users.
The first on depends on AP locations ie a user experiences a higher throughput if the serving AP is closer to the user in terms of attenuation while in the second case, contention among users depends on the active users the serving APs and the channel assignment.

4.4. AP Placement/Localization for Channel Optimiza- tion in FA-WLAN

From equation (1), the objective function measures average throughput per TLP, constraint equation (2) states that a LTP can only be assigned to an AP if the AP is installed, (3) En- sures that each TP is assigned at most once, and equation (4) & (5) limits the number of APs.
In this work, since the existing methods lacks implicit feed- back controls and cannot effectively handle congestion in FA- WLAN under realistic loads, this work therefore developed a layer-4 DMF based on Fuzzy-Logic Adaptive Queuing control- ler (FLAQ) with modulation feedback at the AP base station in order to tune the average queue length and the wireless packet error under realistic load conditions.
The proposed model seeks to maintain buffer size space over a time-varying channel which may exhibit significant degrada- tion in the network bandwidth estimation. Then, the work predicts the lowest packet dropping received by the mobile receiver over a FA-WLAN when there is congestion effect at the AP base station buffer. By using this model, a heuristic TCP performance can be evaluated over a time-varying chan- nel under different conditions of user’s mobility and realistic load conditions. Fig. 4.2 shows a captured AP placement in LTP.

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 183

ISSN 2229-5518

Fig.3c: System Architecture for FA-WLAN

Fig.2: Captured AP Placement in LTP

WL1-Wiredless Link

The proposed TCP layer 4 DMF scenario is shown in Fig. 3a. Fig. 3c shows the system architecture. From Fig. 3b, the con- sidered key preliminaries required to predict the congestion (packet dropping rate, etc) based on our earlier work on fuzzy logic adaptive queuing controller in the FA-WLAN and Ray- leigh fading channel is as follows:
• Multiple TCP flow traffic is considered for all mobile receivers.
• The network is assumed to be stable without heavy or bursty TCP traffic.
• The TCP packet error rate (PER), (i.e., Pr), is caused by the variations of wireless channel when only high- ly bit errors occurs during traffic transmission. As-


WL1-Wired Link

AP Base



suming there is no congestion at the router buffer of
AP base station.
Pr is measured by the channel estimator at mobile re-
ceiver and returned back via the ACK feedback of the round trip of TCP to indicate the sender about the



channel bit errors, so we assumed Pr changes from
5% to 30%.

Fuzzy Logic Controller (Congestion Management) [TCP Layer-4 DMF]

Fig 3a: Unicast-Broadcast FA-WLAN model
Fig.3b: FA-WLAN Process Model
• The TCP rate regulator at the AP router queue of the AP base station is required if and only if multiple TCP flows are present. This rate regulator is mainly used to distinguish the packet error (dropping) due to the variations of wireless channel and the packet loss due to congestion of buffer overflow. In our assumption, there could be packet dropping due to AP buffer overflow is extreme congestion. So, the link could be under or over utilizing bandwidth and the queue threshold of the APs could be exceeded.
At the AP base station (BS), we considered the following as- sumptions:
• Let the buffer size = 256kb packets
• The AP router queue with Qmin = 50kb [packet], and

Qmax =150kb [packet]

• If the average queue length is less than 50kb, no pack-
ets are droppedNo TCP congestion
• If the average queue length is more than 150kb, all the
arriving packets are queued while DMF regulates the
feedback flows to and from the MH or MNs.
• If the average queue length is between Qmin and Qmax, then the packets should be controlled by the fuzzy logic controller depending on to inputs (AVQLcongestion , Pr )
Packet loss rate at AP base station is compensated by DMF

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 184

ISSN 2229-5518

algorithm (hence, there is no packet loss at the event of con- gestion) under realistic loads.


5.1. Simulation Testbed

In this research, the validation of the proposed TCP Modula- tion feedback (L4-DMF) algorithm is considered very vital as this will help to improve congestion management scheme by addressing both packet losses even at busty traffic in a FA- WLAN. In this context, we compared the performance of the developed algorithm (proposed L4-DMF) with that of TCP- TRONVS while establishing the influence of QoS parameters in the developed testbed.
To demonstrate more details on the proposed TCP Modula- tion feedback (L4-DMF) algorithm implementation, a simula- tion testbed was built with OPNET IT Guru 9.1 with DMF con- figured in the OPNET engine as shown in Fig. 4.19, that con- sists of a two wired LAN PC servers (HTTP and FTP/WEB servers), 40mobile client nodes, two AP base stations and an IP gateway router, all connected to the IP cloud. In the mobile nodes including the APs, TCP Tahoe, TCP Reno, TCP NewRe- no, TCP SACK, TCP Vegas as well as the TCP layer-4 Modula- tion feedback (DMF) algorithm were all configure in six inde- pendent scenarios. The subnet sites and IP gateway cloud em- ulates the FA-WAN link with the desired throughput, latency, buffer utilization, data rate stability, queuing length and pack- et loss rate.

5.2. Simulation Configurations

This work implemented Fig. 3c in OPNET IT Guru which was used to generate the parameters for various case scenarios in the simulations. Traffic attributes for the FA-WLAN are listed in Table 4.2. The runtime environment attribute were as fol- lows in the OPNET simulator.
Table 4.2: Basic OPNET Traffic Attributes




Simulation Duration for

Each Scenario

120 mins.


Link Propagation Delays



Switch Output Buffer

100 packets


Simulation Seed



Update Interval

50000 Events


Simulation Kernel



TCP Variants



Mobile clients(Max-min)


Running the test bed, we measured the above metrics achieved by the six TCP versions. Fig. 4.19 shows the valida- tion testbed while Fig 4.20 shows the simulation run/compilation.

Fig.4.19: Validation testbed with FA-WLAN sites_1 and site_2

Fig.4.20: Simulation Plot of TCP-TRONVs with the proposed layer-4 DMF algorithms

5.3. Results and Discussions

Latency Response: In this work, an evaluation on various TCP variants including the TCP SACK was carried out via the sim- ulation model. Fig. 4.21 showed the latency plot of TCP- TRONVs. TCP Vegas and the proposed TCP DMF show a sim- ilarity latency response under realistic loads. The average la- tency for all TCP variants except Vegas and the proposed TCP DMF maintained a fast rise latency throughout the transitions as depicted by the trend curve beginning from 0.00125ms up to 0.00127ms for TCP Tahoe, Reno, NewReno and SACK. In the same vein, the proposed TCP DMF showed a comparative latency response with TCP Vegas (0.0075ms). It maintained a steady rate of about 0.0076ms relative to TCP Vegas. Therefore a difference in the values of latency in the face of equal load per time over the AP controller makes the frame size distribu- tion over TCP variants a major consideration.

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 185

ISSN 2229-5518

Fig.4.21: latency Plot of TCP-TRONVs with the proposed lay- er-4 DMF algorithm

Throughput Response: Fig. 4.22 compares the steady-state throughput response achieved by all TCP variants. Interest- ingly, it was observed that the proposed layer-4 TCP DMF algorithm had slightly better throughput behavior of about

8870 packet/bits compared with TCP Tahoe, Reno, New Reno, SACK and Vegas. This is due to its connection oriented behav- ior leading to accurate fuzzy estimate of the available band- width, buffer size, average queuing length, packet error rate, and fair distribution of frame sizes (packets) under realistic load conditions. As a result of this better throughput behavior, the transmission of realistic traffic witnessed reliable frame data delivery with active connections transmitting data be- tween the mobile nodes and the AP server, with an emulated round trip time equal to 100 ms (a near zero packet loss rates).

Fig. 4.22: A throughput Plot of TCP-TRONVs with the pro- posed layer-4 DMF algorithm

Buffer Utilization Response: Fig. 4.23 shows the buffer utili- zation plot of TCP-TRONVs with the proposed layer-4 DMF algorithm. At the core AP base station and mobile stations, the output buffer was set to hold a maximum of 2048Kbps bytes. Essentially, the buffer sizes were varied starting from 64kb packets to 2028 packets for various realistic load sources. It was observed that with the proposed layer-4 DMF algorithm, there was a fairly assumed peak exponential rise in buffer uti- lization with time due to congestive load effects. Traffic flows shows an enhanced throughput as the buffer sizes were in- creased. The buffer utilization curves are on top of each other with gradual transitions giving way to higher and efficient throughputs. From a buffer size of about 2000kb packets and above, throughput of bulk traffic TCP DMF was significantly

high. This guarantees a fair allocation in the congested link. Thus, an increase in buffer size leads to a better performance network considering the various load intensities. This also means that a better buffer size utilization accounts for less packet drops in a congestive wireless link.

Fig.4.23: Buffer utilization Plot of TCP-TRONVs with the pro- posed layer-4 DMF algorithm

Packet Loss Ratio: Fig. 4.24 depicts a plot of Packet Loss Ratio of TCP-TRONVs with the proposed layer-4 DMF algorithm. As observed from the plot, the proposed layer-4 DMF algo- rithm has a relatively good packet loss ratio as a result of its enhanced throughput behavior while TCP Vegas which has a comparatively better Packet Loss Ratio behavior under con- gestive scenario. Ideally, the TCP Vegas is demonstrates a higher PLR while TCP Tahoe depicts a weak PLR.

Fig. 4.24: Packet Loss Ratio Plot of TCP-TRONVs with the proposed layer-4 DMF algorithm.

Average Queuing length: The proposed TCP_DMF has better Queue management owing its intelligent AVQL prediction. Fig. 4.25 shows the Queuing Length plot of TCP-TRONVs with the proposed layer-4 DMF algorithm. As depicted, the queuing variations with the proposed algorithm show a better spread and congestion control owing to the active queue man- agement (AQM). The TCP Tahoe, Reno, NewReno and SACK on the other hand shows a better management in regions of little or moderate congestion while Vegas and the proposed DMF shows better AVQL management under intense conges- tion considering the preset equilibrium thresholds.

IJSER © 2014

International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 186

ISSN 2229-5518

Fig. 4.25: Queuing Length Plot of TCP-TRONVs with the pro- posed layer-4 DMF algorithm


This research shows that TCP-TRONVS are often unable to give accurate estimates in all FA-WLAN metrics, thereby mak- ing it as a congestion scheme unsuitable for future TCP cloud services. Although there is still room for improvement in TCP performance, the main factor limiting all TCP performance is random packet loss. To overcome this problem, this work propose a new algorithm, layer-4 DMF, that enables TCP to achieve both a higher throughput over wireless links and fair behavior over wired links. The performance of this new algo- rithm has been compared with that of other existing TCP, and it has been found that significant improvement is obtained using the proposed layer-4 DMF taking cognizance the metrics such as Latency, Throughput, Buffer Utilization, and Packet Loss Ratio.


[1] Jerzy Dom˙zał, Nirwan Ansari, Andrzej Jajszczyk, “Congestion Control in Wireless Flow-Aware Networks” IEEE ICC 2011 pro- ceedings, 2011.

[2] Ijaz Haider Naqvi, Tanguy Pérennou, “A DCCP Congestion

Control Mechanism for Wired-cum-Wireless Environments” IEEE Communications Society subject matter experts for publi- cation in the WCNC 2007 proceedings, pp 3915-3920.

[3] http://www.wikipedia/network congestion

[4] Van, Jacobson,Michael J. Karels. Congestion Avoidance and Control (1988).Proceedings of the Sigcomm '88 Symposium, vol.18 (4): pp.314–329. Stanford, CA. August, 1988. This paper origi- nated many of the congestion avoidance algorithms used in TCP/IP.

[5] RFC 2001 - TCP Slow Start, Congestion Avoidance, Fast Re- transmit, and Fast Recovery Algorithms

[6] Jacobson, Van; Karels, Michael (1988). "Congestion Avoidance and Control". ACM SIGCOMM Computer Communication Re- view 25 (1): 157–187.doi:10.1145/205447.205462.

[7] Kurose & Ross 2008, p. 284.

[8] S. Oueslati and J. Roberts, “A new direction for quality of ser- vice: Flow-aware Networking,” in NGI 2005, Rome, Italy, April


[9] A. Kortebi, S. Oueslati, and J. Roberts, “Cross-protect: implicit

IJSER © 2014

service differentiation and admission control,” in IEEE HPSR

2004, Phoenix, USA, April 2004.

[10] “Implicit Service Differentiation using Deficit Round Robin,” in

ITC19, Beijing, China, August/September 2005.

[11] J. Domzal and A. Jajszczyk, “Approximate Flow-Aware Net- working,” in IEEE ICC 2009, Dresden, Germany, June 2009.

[12] A. Kortebi, L. Muscariello, S. Oueslati, and J. Roberts, “On the

scalability of fair queueing,” in ACM HotNets-III, San Diego, USA, November 2004

[13] W. Richard Stevens. TCP/IP Illustrated, Volume 1: The Protocols.

Addison Wesley,1994


[15] R. Bruno, M. Conti, and E. Gregori, “Analytical Modeling of TCP Clients in Wi-Fi Hot Spots”, Proceedings of IFIP Networking Conference, Athens, Greece, pp. 626-637, May 2004.

[16] A. Akella, S. Seshan, and A. Shaikh, “An Empirical Evaluation of Wide-Area Internet Bottlenecks,” Proc. of ACM Internet Measurement Conf. (IMC), (2003).

[17] M. Hirabaru, “Impact of Bottleneck Queue Size on TCP Proto-

cols and Its Measurement,” (2006), IEICE Trans. on Information and Systems, vol E89-D, no.1, pp. 132-138.

[18] S. C. F. Chan and J. Y. B. Lee, (2011), “A Novel Link Buffer Size Estimation Algorithm for Bandwidth Varying Mobile Data Networks,” Proc. of IEEE 7th Inter. Conf. on Wireless and Mobile Computing, Networking and Communications (WiMob).

[19] C. P. Fu and S. C. Liew, (2003), “TCP Veno: TCP Enhancement for Wireless Access Networks,” IEEE J. of Selected Areas in Com- munications, Vol.21, No.2.

[20] H. Abdel-Jaber, M. Mahafzah, F. Thabtah, and M. Woodward, (2008), “Fuzzy Logic Controller of Random Early Detection based on Average Queue Length and Packet Loss Rate”, Proc. Of SPECTS 2008.

[21] J. D. Mallapur, S. S. Mavi, and D. H. Rao, (2010), “Fuzzy-Based Approach for Packet Dropping in Wireless Networks,” Int. J. of Advanced Networking and Applications, vol. 1, no. 5, pp. 301-306.

[22] A. E. Abharian and M. Alireza, “Hybrid GA-BF Based Intelli- gent PID Active Queue Management Control Design for TCP Network,” IEEE Inter. Conference, (2011),

[23] K. Zhou, K. L., Yeung, and V. O.K. Li, “Nonlinear RED: A Sim- ple Yet Efficient Active Queue Management Scheme,” Comput- er Networks, vol. 50, pp. 3784-3794, (2006),

[24] A. Athuraliya, S. H. Low, V.H. Li, et al., (2001), “REM: Active

Queue Management,” IEEE Network, vol. 15, pp. 48-53.

[25] T. Shikama, (2010), “Mitigation of Bursty Packets By a TCP Proxy Improving TCP Performance in a Wired and Wireless Network,” IEEE Globcom Workshop On Complex and Com- munication Networks.

[26] P. Dalal, N. Kothari and K.S. Dasgupta, (2011), “Improving TCP Performance over Wireless Network with Frequent Disconnec- tions,” International J. of Computer Networks & Communica- tions (IJCNC), vol. 3, no.6, pp. 169-184.

[27] M. Miyoshi M. Sugano, and M Murata, (2002), “Improving TCP Performance for Wireless Cellular Networks by Adaptive FEC Combined with Explicit Loss Notification,” IEICE Trans. Comm. VOL-E85, no. 10, pp.2208-2213.