Conference

advertisement
CLUSTERING APPROACH TO PREDICT THE CONGESTION
LEVEL IN A WIRELESS NETWORK
Pravalika D
U.G Student, Department of ECE
Sridevi Women’s Engg. College
Hyderabad, Telangana
pravalikadonti@gmail.com
Eesha L
U.G Student, Department of ECE
Sridevi Women’s Engg. College
Hyderabad, Telangana
eeshalella7@gmail.com
Mounika M
U.G Student, Department of ECE
Sridevi Women’s Engg. College
Hyderabad, Telangana
madagonimounika14@gmail.com
Dr.V. Balaji
Professor, Department of ECE
Sridevi Women’s Engg. College
Hyderabad, Telangana
balaji.phd.auc2008@gmail.com
Abstract— Congestion in a network is due to the lower
bandwidth in the wireless part as compared to the wired one.
Extensive planning has to be made on the wireless network side
as it is challenging to predict the number of nodes which are
connected to the network over a period of time. In this paper
we study the link between the Re Transmission Timer with
respect to network congestion in a network. We are proposing
a cluster based methodology to determine the status of the of
access point with respect to the channel which will help us to
plan better network. We use the traces collected from wireless
monitoring at the 62nd Internet Engineering Task Force
(IETF) meeting held in Minneapolis, MN, March, 2005
network is typically well provisioned to handle the network
load. Therefore, there arises a compelling need to
understand the performance
of the wireless portion of
heavily utilized and congested wireless networks[2]. The
evaluation of wireless network requires the generation of
workloads to test the capability and performance of the new
protocol or technique being studied. Lack of realistic traffic
is a major limitation that has forced researchers to generate
synthetic data for their simulations and experiments to
evaluate performance of wireless technologies.
Keywords— Congestion, TCP flags, Wireless monitoring,
Clustering, Access point.
The rest of the paper is organized as follows. In Section II
we present motivation and related work. Our experimental
and analytical methodology is presented in Section III,
Finally in Section IV, we conclude with a discussion of how
this research may be applied to solve the current issues in
the wireless networks.
II. MOTIVATION
I. INTRODUCTION
Wireless network is a type of network which
communicates using interconnections through the nodes
without the use of any wires [1]. Its technology is
implemented with the help of electromagnetic waves and its
implementation occurs at physical layer. One of the
important aspects of the network is in expensive in nature. A
wireless network allows devices to stay connected to the
network but roam untethered to any wires. Access points
amplify Wi-Fi signals, so that the device can be far from a
router but still be connected to the network
Wireless networks also use the Open System Interconnect
(OSI) reference model in the transmission of data. The
manner in which this reference model applies to wireless
networks is similar to wired networks with some differences
in the data link layer where wireless networks coordinate
access by data to a common air medium and also deal with
errors which occur due to the inherent nature of the wireless
medium. At the Physical layer, the data is transmitted in the
form of radio waves.
Congestion control is a network layer issue, and is
thus concerned that what happens when there is more data in
the network that can be sent with reasonable packets delay,
no lost packets .The effect of multiple losses in one RTT in
TCP transmission of a typical proactive scheme and a
typical reactive scheme in the wireless environment.
Wireless network with a high density of nodes and within a
single collision domain has a high probability of congestion,
decreasing the performance significantly. In general
congestion include drastic drops in network throughput,
unacceptable packet delays and session disruptions.
Typically the back-haul wire line portion of a wireless
XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE
The TCP is widely used in wired network and the result
of congestion is any packet loss and hence the congestion
window size is reduced. Fading, shadowing and hand off are
some of the losses that occur infrequently in wireless links.
M. Balazinska et al [4], D. Kotz et al[5], T. Henderson et
al[6] A. Jardosh et al[7] and V. Balaji et.al [8] have carried
out various studies and research in wireless network
deployments. A wide range of wireless network behavior is
analyzed in these studies, which provide insights into the
actions of deployed networks in different scenario. Majority
of the studies have been conducted within the wireless LAN
of university campuses. Their research includes extensive
amounts of raw wireless traffic data which has been
collected for subsequent analysis and research. A proposal
to create a model for congestion prediction is made using
such raw data. Real time data is chosen because it is not
only limited to simulations but can also be used for
experimental deployments which is essential for wireless
protocol evaluation.
TCP congestion control algorithms plays a critical role in
improving the performance of TCP and by preventing
congestion collapse we can regulate the amount of network
traffic on the internet. However, it is a challenging task to
predict whether a complex network has a normal behavior
or not and analyze network dynamics. One of the most
important elements of TCP sender state that can help us
study the features of TCP per-connection states in the
Internet is congestion window.
Wireless local area networks (WLANs) have become
very popular, but the complex behavior of wireless signal
propagation creates significant challenges. We present an
efficient opportunistic retransmission protocol
that
improves network performance in dynamic infrastructure
WLANs. The idea is to exploit overhearing nodes to
retransmit (or relay) on behalf of the source after they learn
about a failed transmission. Opportunistic retransmission
leverages the fact that wireless networks innately use
broadcast transmission and that errors are mostly location
dependent. Thus, if the intended recipient does not receive
the packet, other nodes may have received the packet and
thus become candidate re transmitters for that packet. With
multiple wireless devices distributed in space, the chance
that at least one available device can transmit the packet
increases. Candidate relays participate if they have a higher
chance of delivering the packet successfully than the source,
thus results in an increased throughput.
The area of RTO estimation is an area of TCP that
has not received the same level of analysis as other TCP
flow control mechanisms. Estimating a suitable value for the
RTO is very important. Too small a value may result in
needless sender time out regardless of the ACK being in
transit from receiver to sender. TOO large and RTO value
could result in significantly reduced overall excellent flow.
Recent results from a particular large scale internet traffic
study by Balakrishnan et.al[BP98] have shown that
approximately 50% of all packets losses required a time out
to recover. another recent study found that over 85% of all
time outs are due to non trigger of the fast retransmit
mechanism while there are some proposals that look to
reduce TCP’S reliance on time out expiry, these will take
time to be discusses, agreed-upon an perhaps ultimately
deployed on a wide enough scale in the mean time, there is
clear need for renewed research into the TCPRTO
mechanisms.
III. GOAL OF OUR WORK
In this paper we investigate raw wireless trace files and the
development of a clustering model, the analysis of lost
packet segment has been done for the networks which are
connected using Wi-Fi by creating a cluster model among
the nodes connected together.
Majority of the TCP segments transmit data while
others are simple acknowledgements for previously received
data. Using the 3-way handshake, it completes the
connection before data is transferred. The purpose of each
segment in TCP is resolute with the help of the TCP flag
options. It is a control bit that indicates different connection
states and information about how a packet should be
handled. This facilitates both transmitter and receiver to
specify the flags to be used so that data handling is correct.
Acknowledgment flag is used for successful delivery of
packets. Loss segment packet is the foundation of our work.
For implementing our methodology data from a wireless
network deployed at Internet Engineering Task Force
(IETF) on March 2005 was used. The IETF network
consisted of 38Airespace2 1250 access points (APs)
distributed over three floors. Each Airspace Access Point
supported up to four virtual APs. A virtual AP is a logical
AP that exists within a physical device and enables the
wireless LAN to be segmented into multiple broadcast
domains. Thus, at the IETF, a total of 112APs (38 physical
APs x 4 ESSIDs per physical AP) were available for
utilization [8]. In this work we study the data over a 60
minute interval and consider only the TCP and UDP packets
as compared to the evaluation methods used by[8].
Figure 1 shows the TCP and UDP traffic flow captured
during 60 minutes.
Figure 2 shows the duplicate acknowledgement received
due to packet loss.
The Multi Step cluster method we propose is a
scalable cluster analysis algorithm designed to hold very
large data sets. In the first pass the pre clustered data is
converted to smaller clusters. In the second pass the smaller
clusters are broken down to still smaller clusters. The precluster step uses a sequential clustering approach. It scans
the data and records one by one and decides if the current
documentation should be merged with the previously
formed clusters or starts a new cluster based on the distance
criterion mentioned herein. The procedure is implemented
by constructing a modified cluster feature (CF) tree. The CF
tree consists of levels of nodes, and each node contains a
number of entries. A leaf entry (an entry in the leaf node)
represents a final sub-cluster. The non-leaf nodes and their
entries are used to guide a new record quickly into a correct
leaf node. If the CF tree grows beyond allowed maximum
size, the CF tree is rebuilt based on the existing CF tree by
increasing the threshold distance criterion. The rebuilt CF
tree is smaller and hence has space for new input records.
A set of observation are assigned into smaller groups called
clusters. This process is known as Clustering whereby the
observations in the same cluster are related in some sense.
Being an unsubstantiated learning method, for statistical
data analysis and in machine learning, data mining, pattern
recognition, image analysis and bioinformatics a clustering
method is often used.
The log based distance measure is the most popular
method for measuring the distance between clusters. The
distance between two clusters is related to the decrease in
log possibility as they are combined into one cluster. The
distance between clusters j and s is defined in the section
below.
REFERENCES
[1]"Overview
of
Wireless
Communications".
cambridge.org.
http://www.cambridge.org/us/catalogue/catalogue.asp?isbn=
0521837162&ss=exc. Retrieved 2008-02-08.
[2] Amit P Jardosh, Krishna N Ramachandran, Kevin C
Almeroth, Elizabeth M Belding-Royer “Understanding
Congestion in IEEE802.11b Wireless Networks” University
of California.
[3] Stefan Karpinski, Elizabeth M. Belding, Kevin C.
Almeroth “Towards Realistic Models of Wireless
Workload”, University of California.
[4] M. Balazinska and P. Castro, “Characterizing mobility
and Network usage in a corporate wireless local-area
network,” in ACM MobiSys, San Francisco, CA, USA, May
2003, pp. 303–316.
[5] D. Kotz and K. Essien, “Analysis of a campus-wide
wireless network,” in ACM MobiCom, September 2002.
[6] T. Henderson, D. Kotz, and I. Abyzov, “The changing
usage of a mature campus-wide wireless network,” in ACM
MobiCom, September 2004.
[7] A. Jardosh, K. Ramachandran, K. Almeroth, and E.
Belding-Royer, “Understanding link-layer behavior in
highly congested IEEE 802.11b wireless networks,” in
ACM Sigcomm EWIND, Philadelphia, PA,USA, August
2005.
The clustering result obtained for the nodes connected using
wireless access point is shown in the figure 3. It is observed
that there was 33% of abnormal loss occurred when the
packets are exchanged between the nodes in a wireless
connection.
[8] V.Balaji ,V.Duraisamy “cluster based packet loss
prediction using TCP ACK packets in wireless network”
[8] http://hubpages.com/hub/congestion-control
[9]
http://en.wikipedia.org/wiki/TCP_congestion_avoidance_al
gorithm 105
[10] Balakrishan H, Padmanabhan V, Seshan S,Stemm M
and Katz M,”TCP Behaviour of a Busy Internet Server.
Analysis and Improvements”, Proceedings 0f INFOCOMM
98,San Francisco, march 1998
IV CONCLUSION
We proposed a multi-step clustering algorithm for clustering
out the time periods where an abnormally high packet loss
has been discovered due to either congestion or other effects
common in wireless networks. The clustering algorithm
proposed was able to cluster the data with very good
efficiency. Further studies need to be done for a longer time
frame.
Download