Professional Documents
Culture Documents
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 53
Abstract—Wireless networks and communications made revolution into human life. The demand for wireless networks is increasing and
wireless networks need to meet lot of challenges. Once such issue if design of efficient MAC protocol for wireless networks. In this paper, we
study and develop a new MAC protocol based on the concepts of pipelining techniques. We observe that the performance of the proposed
protocol is good.
Index Terms—Wireless Network – MAC – pipelining – control channels – data channels etc.
—————————— ——————————
this thesis, we focus on the issue of medium access control in
1 INTRODUCTION wireless networks. The function of medium access control
1.1 Wireless Networks (MAC) is to enable the effective sharing of the channel among
all nodes in the network. Ideally, for both wired and wireless
Wireless networks are fast becoming popular as they allow
networks, the MAC protocol should facilitate perfect channel
users the ability to remain connected while on the move. These
utilization while still being fair to all flows within the network.
wireless networks can either be infrastructure‐based or ad hoc
The conflict between these goals usually does not allow the
networks. Infrastructure‐based networks involve wireless
ideal to be realized in practice. In wireless networks MAC
devices communicating with an access point (AP), which then
protocols can be classified either as reservation‐based protocols
connects to the Internet via a wired network. On the other
or contention‐based protocols. Reservation‐based protocols are
hand, ad hoc networks are a collection of devices outfitted with
used in situations where an access point is present. The AP
wireless transceivers that enable communication between the
polls the nodes before assigning access rights in turn and a
nodes, independent of any pre‐existing infrastructure. This
node is only allowed to transmit when it is allocated the right
independence is a major benefit as it enables communication is
to do so. These protocols are also used to provide quality of
situations where there is no time to set up the necessary
service (QoS) guarantees for realtime traffic. An example of a
infrastructure, e.g., military and rescue operations, or in
reservation‐based protocol is the IEEE 802.11 Point
situations where the need of a communication network is
Coordination Function (PCF). Contention‐based protocols are
temporary, e.g., conferences. Wireless networks face a
distributed algorithms that allow nodes in an ad hoc network
multitude of challenges that the traditional wired networks do
to be able to communicate effectively over the shared channel
not. Wired networks typically communicate over wires made
and are largely based on Carrier Sense Multiple Access with
from copper or fiber optics, where the probability of error is
Collision Avoidance (CSMA/CA).
very small. Collision detection is generally fast since nodes are
able to listen to the medium while they are transmitting. The 1.2 Medium Access Control (MAC)
topology of a wired network is usually fixed, which greatly New Medium Access Control (MAC) protocols are constantly
simplifies routing. Furthermore, wired networks have being developed in an attempt to approximate the ideal as
dedicated routers whose sole function is to ensure that the closely as possible. Some protocols optimize the system
packets are correctly routed form source to destination. In throughput at the expense of fairness while others give
contrast, wireless networks communicate over a highly error preference to achieving a greater level of fairness in wireless
prone medium. Collisions become harder to detect since the networks. Given that protocols are continually being proposed,
wireless transceivers cannot receive reliable while transmitted various metrics have also been developed in order to be able
on the same channel. As such, the absence of a reply is usually to evaluate the efficiency and robustness of these protocols.
the only indication that a collision has occurred and the time One common metric is channel utilization, which is defined to be
required for detection is dependent on the length of the the fraction of time spent on successful transmissions. Another
transmitted packet. The topology of an ad hoc network is metric is saturation throughput, which represents the maximum
constantly changing and node membership within the network throughput the system can achieve under stable conditions. Both
is also not fixed. This makes neighbor discovery and routing channel utilization and saturation throughput are directly
nontrivial. Furthermore, network connectivity is dependent on affected by the amount of overhead a protocol introduces for
nodes voluntarily forwarding packets for each other, in order each successful transmission and the efficiency with which
to route a packet from source to destination. These are just collisions are resolved. Protocol fairness is also an important
some of the many challenges that wireless networks face. In issue. The function of Medium Access Control (MAC) is to enable
———————————————— the effective sharing of the channel among all nodes in the
V. Kumar is with the Department of Computer Science and Engineering, network. Ideally, for both wired and wireless networks, the
Knowledge Institute of Technology, Salem, Tamilnadu, India. MAC protocol should facilitate perfect channel utilization while
P. Balasubramanie is with Department of Computer Science and Engineering, still being fair to all flows within the network. The conflict
Kongu engineering College, Perundurai, Tamilnadu, India.
between these goals usually does not allow the ideal to be
realized in practice.
54
We will present some of the research that has been done in idle slots within a transmission cycle while the bandwidth‐
the field of Medium Access Control (MAC) for ad hoc dependent component accounts for the collisions that
networks. Much of the research here is motivated by the fact occur. In terms of these components, analytically
that the prior research done with respect to wired networks is compares the capacity of the generic un‐pipelined MAC
not directly applicable to the wireless arena. This is a direct protocol with that of the generic pipelined MAC protocol.
consequence of the different properties of the wired and Their analysis shows that the pipelined protocol cannot do
wireless medium. any better than the un‐pipelined protocol if there is no
One of the earliest MAC protocols proposed was ALOHA bandwidth‐independent component in the transmission
(Yang and Vaidya, 2003a). In pure ALOHA, nodes transmit cycle. On the other hand, if there are no bandwidth‐
whenever they have packets to send, without regard to the dependent components in the transmission cycle,
current state of the medium. A positive acknowledgment pipelining does lead to a gain in performance. The
scheme is used to determine if the transmission was performance gain decreases as the ratio of the idle
successful. This is a common feature in most MAC protocols overhead to the data packet size increases. Thus the
designed for the wireless medium since the wireless transceiver benefit of pipelining will depend on how much of the
does not allow a node to transmit and listen to the channel transmission cycle is bandwidth‐dependent and how much
simultaneously. If no ACK returns, a collision is assumed to is bandwidth‐independent.
have occurred and the node retransmits after a random delay. Yang and Vaidya [2] propose Pipelined Packet Scheduling
Since no form of carrier sensing is done, a transmitted packet is (PPS) that partially pipelines the contention resolution
vulnerable to collision for an interval equal to twice its process over a data channel and a narrow busy tone
transmission time. Put differently, if a second transmission starts channel. The contention resolution process is divided into
within a packetʹs vulnerable period, a collision occurs and both two stages and both stages adopt the back‐off algorithm,
transmissions are unsuccessful. A throughput analysis shows each with its own set of CWmin and CWmax values. All
that pure ALOHA utilizes at most 18% of the available channel backlogged stations start in the first stage and decrement
bandwidth. A simple extension to pure ALOHA was to divide the back‐off counter every slot, regardless of the channel
time into discrete intervals, known as slots and allow nodes to status. When one nodeʹs back‐off reaches 0, it transmits a
only transmit at the beginning of a slot. This extension, known busy tone to inform its neighbors that it has won the first
as slotted ALOHA, halves the vulnerable period of a packet and stage. All nodes that hear this busy tone remain frozen in
is able to achieve a maximum channel utilization of 36%, i.e., stage 1 until the next transmission commences. While the
double that of pure ALOHA. first stage occurs in parallel to the data transmission, the
To further increase the channel utilization, Carrier Sense second stage begins at the end of the current transmission
Multiple Access (CSMA) protocols were developed (Yang cycle. All nodes that enter the second stage wait for a
and Vaidya, 2003b). CSMA protocols can be divided into second back‐off interval before transmitting. If the
non‐persistent CSMA and p‐persistent CSMA. In both channel remains idle until the end of the back‐off duration,
categories, nodes first sense the channel before any action is the node transmits its packet. If the channel is sensed busy
taken. In non‐persistent CSMA, the node transmits if the before the end of the back‐off interval, the node loses the
medium is idle. If the medium is sensed as busy, the node waits contention, returns to the first stage, and recon tends. If
a random delay period before retrying. In p‐persistent CSMA, collisions occur in the second stage, the contention
the node continually senses the medium until it is idle. It then windows in both stages are exponentially increased. The
transmits in a give slot with probability p an defers end of the second stage is marked by a successful
transmission to the next slot with probability 1‐p. If a collision transmission. In this way, much of the protocolʹs idle
occurs, the node waits a random delay before retransmitting. In overhead is effectively masked. The authors show that, at
the limit, 1‐persistent CSMA employs the greedy principle and high loads, unlike IEEE 802.11, the saturation throughput
always transmits as soon as it senses the channel as idle. remains close to the peak value. PPS also maintains the
Crow et al. (1997), Yan and Park (2004) and Bianchi et al. same level of fairness IEEE 802.11 displays. Further analysis
(2006) shows that the CSMA protocols are superior to both shows that the performance achieved is not sensitive to the
pure and slotted ALOHA. CSMA protocols not only achieve probability of busy tone detection. In fact, even when a
significantly higher throughputs, they also perform better busy tone is not transmitted, the performance of PPS does
under high network loads. CSMA is also more efficient than not degrade significantly. Using this fact, the authors extend
ALOHA as it has smaller delay times and requires a smaller the PPS algorithm in [] to propose the Dual Stage Contention
number of retransmissions per successful transmission. Given Resolution (DSCR) protocol. DSCR has the same protocol
these obvious benefits of incorporating carrier sensing into the structure but does away with the need for the busy tone
MAC protocols, most of the succeeding protocols have been channel. In DSCR, all deferring nodes, excluding the
based on the CSMA model. node that just transmitted the successful packet,
decrement their first stage back‐off counter by a fixed value
F at the end of a successful packet transmission. After
2 ANALYSIS
this, nodes decrement their back‐off counters for each
The concept of pipelining in MAC protocols is not a new idle slot sensed. Notes enter stage 2 when their back‐off
one. Todd and Mark [14] consider this issue for a generic counter in stage 1 is less than or equal to 0. If nodes enter
MAC protocol. They divide each transmission cycle into a stage 2 at the beginning of transmission cycle, they wait a
bandwidth‐independent component, a bandwidth‐ random back‐off interval before attempting to transmit. If a
dependent component, and the successful transmission. node enters stage 2 during a transmission cycle, it
The bandwidth‐independent component accounts for the immediately commences transmission. Once a
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151‐9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 55
receiver switch over to the data channel to do the actual
Data/ACK transmission. This protocol adapts the IEEE
802.11 MAC protocol to explicitly support pipelining.
The protocol also assumes that each node has one
transceiver and hence is only able to transmit or receive
over one channel at any instant. Figure 2 illustrates how the
principle of pipelining is employed across the control and
data channels. As seen from the figure, while the Data1 is
being transmitted on the data channel, the remaining nodes
stay on the control channel and commence the contentions
(RTS2/CTS2) for the next data packet, Data2 that will be
Fig. 1 Basic Pipelining
transmitted once the Data1/ACK1 dialog has completed.
Thus, while one data transmission is ongoing on the data
transmission commences on the channel, the other
channel, contention resolution for the next data packet is
contending nodes return to stage 1, double their CW sizes,
simultaneously occurring on the control channel. In this
and re‐contend. Simulation results in [16] show that DSCR
way, the protocol is able to effectively mask the overhead
achieves a higher performance than IEEE 802.11 and
generated by the contention‐resolution process. Ideally,
MACAW, even in the presence of hidden terminals. DSCR
each contention‐resolution cycle should complete before the
also uses a smaller minimum value for the contention
current Data/ACK transmission on the data channel
window in stage 2. This increases the difference in CW sizes
completes so as to ensure maximum utilization of the data
between the winning node and the rest of the network,
channel.
similar to the situation seen in [7]. As a result, DSCR
The protocol is currently designed for WLANs. As such,
performs worse than IEEE 802.11 with respect to fairness.
[16] Show that if the value of F is adaptively modified, so the following descriptions are based on an isolated
that nodes that have been in stage 1 longer decrease their WLAN, where all nodes are within the transmission
backoff counters at a faster rate, the fairness of DSCR is range of each other.
significantly improved, though the performance gain is
3.1 Definition of Terms
reduced.
Yang and Vaidya [17] compare PPS, DSCR, and the We define the following terms for ease of exposition; they are
explicit pipelining scheme, similar to DCMAC. As the further illustrated in Figure
authors noted, the explicit pipelining scheme is only Contention resolution cycle, W — This is the interval from the
beneficial if both stages in the pipeline are balanced. This time the current contention began until the successful node
balance may be too difficult to achieve when pipelining switch to the data channel to commence data
IEEE 802.11 as the performance is expected to be heavily transmission
dependent on the bandwidth division, the distribution of Data transmission cycle, T — This is the duration of the
the packet size, and the number of flows. Here, we current Data/ACK transmission on the data channel.
implement an explicit pipelining scheme and evaluate the Start_time ‐ This state variable denotes the time the
exact dependence of the performance on these factors. This current data transmission cycle began.
will enable us to conclusively state if pipelining IEEE 802.11 is End_Time ‐ This state variable either contains the time
as impractical as it intuitively seem. that the current data transmission cycle will end or denotes
In this paper we attempt to improve the performance the time the most recent data transmission cycle ended.
achieved by masking the contention overhead via
Current_sender, Current_receiver ‐ These state variables
pipelining. Pipelining allows the execution of various jobs to
maintain the addresses of the sending and receiving nodes,
overlap in time. This is accomplished by dividing each job
respectively, that are currently on the data channel. All
into stages, and as soon as the first job finishes a stage, the
nodes within the network maintain these four state
second job enters that stage as the first job moves on to the
variables so as to enable the correct execution of the MAC
next stage. Figure 1 illustrates this concept by using a three‐
protocol.
stage pipeline. In MAC protocols, the ʹjobʹ is the dialog to
transmit a packet and it is divided into a contention‐ Given that contention resolution and the data
resolution stage and a packet transmission stage. transmission occur on different channels, the duration for
each cycle has to be advertised. Thus, both the RTS and the
CTS control packets have an extra duration field added to
3 PROPOSED TECHNIQUE them. The requestTime duration field advertises the interval
The proposed protocol uses the concept of pipelining to that this sender and receiver pair expects to wait before
mask most of the overhead generated by the contention switching to the data channel to begin transmission. The
resolution process, so as to enhance the overall channel second duration field, dataTime, holds the duration of the
utilization in a WLAN. The total bandwidth is divided corresponding Data/ACK exchange on the data channel.
into a control channel and a data channel. This protocol is For example, from Figure 2, the packet RTS2 will have its
divided into a contention‐resolution stage and a data requestTime field set to t4‐t2while CTS2 will have its
transmission stage. Contention resolution, via the use of the requestTime field set to t 4‐t 3 . Both RTS2 and CTS2 will have
RTS/CTS mechanism, is performed over the control channel. their dataTime fields set to t5‐t4.
Upon winning the right to transmit, the sender and the
56
All nodes start on the control channel. Any node that
wants to transmit must first carrier‐sense the control
channel. If the channel is sensed to be busy, the node
defers its transmission until the channel is idle for at least a
DIFS duration. Once the channel is sensed to be idle for at
least DIFS, the node randomly selects a back‐off counter
value, b, from the uniform distribution over the interval
[0,CW‐1]. Similar to IEEE 802.11 DCF, a node must wait for b
idle slots before it is allowed to transmit.
Once the back‐off counter reaches 0, the node, say S,
transmits an RTS to its destination node, say R, over the
control channel. The requestTime field in the RTS
contains the duration t1 that S expects to wait before it is
able to switch to the data channel and begin transmission.
The minimum value of t1 is T=SIFS+CTS_Time+2*SIFS, which
is the smallest duration that is required for the contention‐
resolution to transmit the CTS over the control channel.
When R receives the RTS, it replies with a CTS, after an SIFS Fig. 2: MAC with Different Channels
interval. R could have had a different duration t2 than it
data transmission commences. Otherwise the nodes enter a
had expected to wait before switching to the data channel.
busywait loop, where they continually sense the channel and
So when replying, the requestTime field in Rʹs CTS is
begin transmitting only when the channel is sensed idle.
composed as follows:
Since we are currently working within an isolated WLAN,
CTS^requestTime=max(tl ‐ CTS_Time ‐ SIFS, t2) When S
the assumption is that both the sender and the receiver
receives the CTS, it will wait the duration that R had stated
should see the same channel conditions. This assumption
in the CTS. This will ensure that both nodes S and R switch
does not always hold, and the impact of differing channel
over to the data channel simultaneously.
conditions is discussed in a later section. Given that both S
Nodes other than S and R that overhear the RTS or the
and R had maintained a state that was consistent with the
CTS use the duration fields and the address fields to update
system, the NAV of the other nodes would expire at the same
their ovwn state. The address fields are used to update the
time S begins transmission on the data channel. At this point,
Current_sender and Current_receiver fields while the
these nodes can resume decrementing their back‐off and
nodes set their NAV on the control channel for the duration
contend for access rights for the next data transmission cycle.
indicated in the requstTime field. This is to prevent
We illustrate this protocol with an example. There is a
collisions on the control channel, as nodes defer their
flow from node 1 to node 2 and from node 3 to node 4. Also
transmission when the NAV is set. The NAV is also used to
assume that both nodes 2 And 4 have backlogged packets to
prevent more than one flow within reception range of the
other nodes not shown in the Figure. As seen from Figure 2,
RTS or CTS for attempting to reserve the data channel for
node 1 is the first to decrement its back‐off counter to 0 and
the next data transmission cycle.
hence sends an RTS to node 2. Upon sensing this
Once an RTS and CTS have been exchanged on the
transmission, all the other nodes freeze their back‐off
control channel, it implies that the data channel has been
counters. When node 2 receives the RTS, it replies with
reserved for the next transmission. As such, none of the other
CTS. Since there is an ongoing data transmission, node1
nodes is allowed to resume contending on the control
and 2 set their requestTime fields to last until the end of the
channel until the next data transmission cycle commences.
current data transmission, plus the extra duration needed to
Thus, by setting the NAV to expire only when S and R
sense the data channel. Nodes 3 and 4 overhear both
switch to the data channel, the occurrence of multiple
control packets and update their NAVs accordingly. At
reservations is also prevented. The nodes also update
the end of the current data transmission cycle, nodes 1
Start_time to reflect the end of their NAV and use the
and 2 switch to the data channel and sense the medium.
dataTime field in the RTS or CTS to compute End_time,
Since it is sensed as idle, the data transmission commences.
both of which are used to track the transmission on the data
Al the same time, the NAVs on nodes 3 and 4 expire and
channel. The updating of the Start_Time and End_Time is
they resume decrementing their back‐offs. The back‐off
done as follows: Start_Time=Time at which the NAV expires
counter at node 2 remains frozen at its current value
End_Time=Strat_Time+dataTime If there already is an
throughout the entire data transmission cycle as it is currently
ongoing transmission on the data channel, these state
the receiver on the data channel.
variables are assigned their new values at the end of the
The common receiver case requires special attention.
current data transmission cycle. This ensures that the state
The common receiver scenario occurs when at least two
variables Start_time, End_time, Current_receiver, and
sources are sending packets to the same destination node.
Current_sender are always consistent with the current data
For example, both nodes 2 and 4 are sending packets to a
transmission.
node 5. In this case, if the flow from node 4 to node 5 won the
Assuming that nodes S and R win the contention on the
channel access, both nodes will move to the data channel to
control channel, they will switch to the data channel at the
commence transmission. Once node 4 starts transmitting,
end of the current data transmission cycle. Instead of
contention‐resolution also resumes on the control channel. In
transmitting immediately, both nodes first carrier‐sense
the case where node 2 does not remember that node 5 is
the data channel. If the channel is sensed to be idle, the
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151‐9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 57
currently on the data channel, if it is the first to decrement its channel to be busy. So if the interference source continually
back‐off value to 0, it will send out an RTS for node 5. Since affects only a fixed set of nodes, then the rest of the node have
node 5 is on the data channel, it will neither receive nor be a higher probability of winning the contention. This is true
able to respond to node 2ʹs CTS on the control channel. As assuming that the destination of the unaffected node also does
such, node 2 will not receive a reply to its RTS. In accordance not sense the interference.
with the protocol, the absence of a reply is taken to On the data channel, differing channel conditions could
indicate that a collision occurred. This will cause node 2 to cause either the sender or receiver to enter the busy‐wait loop
unnecessarily double its contention window and retry after while the other node either commences transmission or expects
a new random back‐off delay. Consequently, the flow from an incoming transmission. If the sender enters the busy‐wait
node 4 to 5 will be able to capture the channel for an loop, the receiver could time out and return to the control
interval before node 2 is able to win the channel access. In channel before the sender is able to commence transmission.
the worst case, this could lead to an exponential increase the As such, the data transmission will be unsuccessful and the
size of the nodeʹs CW and may be even starve the flow from 2 sender will have to return to the control channel and initiate
to 5. Once node 2 wins the channel, it captures the channel the retransmission process. If the receiver enters the busy‐wait
until node 4 is able to steal it back. Fortunately, this loop, it will still be able to receive the sender’s transmission.
phenomenon is easily prevented by using available Depending on how strong the interference is, the packet may
information. Since all nodes maintain the state variables or may not have been corrupted. If the packet is not corrupted,
Current_sender and Current_receiver, if the address of a then the transmission will be successful as no carrier sensing is
nodeʹs destination is equal to the address in either of these done before the transmission of an ACK. If the packet was
variables, the node freezes its back‐off counter until this corrupted, the receiver does not reply with an ACK and
data transmission has finished before resuming its back‐ returns to the control channel. Consequently, it is still possible
off. This not only prevents the exponential increase in the for the transmission to be unsuccessful and result in
size of a nodeʹs CW, it also increases fairness between flows retransmission.
as channel capture by any individual flow is reduced. This can also be extended to multi‐hop networks. The
As with any MAC protocol, collisions have to be spatial distribution of the various flows generally results in
accounted for. On the control channel, collisions occur varying channel conditions. For the case of the WLAN, the
when two nodes transmit an RTS at the same time. As in interference was a result of an external source situated at close
IEEE 802.11, collisions are detected by the absence of CTS. proximity and affected a certain subset of the nodes. In a
When a collision is detected on the control channel, the multi‐hop network, the flows themselves act as the sources of
colliding nodes double their contention windows and pick a interference for other flows. As such, the channel conditions
new random back‐off value. The other nodes that detect the across the network are constantly varying and dependent on
collision set their NAV to EIFS. Since the EIFS duration is the state of each flow. This also gives rise to the hidden‐
used to protect impending CTS, the value of EIFS has been terminal problem which results from the sender’s inability to
modified to reflect the time needed for CTS to be successfully detect the channel conditions at the receiver. Thus, the severity
transmitted. If a collision occurs on the data channel, as a of the performance loss would be worse in multi‐hop
result of differing channel conditions, the colliding nodes networks.
have to re‐contend for the access rights and hence return to Thus, differing channel conditions could seriously degrade
the control channel, double their CW, and pick a new back‐ the throughput achieved by increasing the duration of the
off value. average contention‐resolution cycle and by increasing the
Unlike the IEEE 802.11, this protocol only works with the number of unsuccessful transmissions on the data channel.
RTS/CTS access method. All data transmissions must be The general performance characteristics for this protocol are
precluded by and RTS/CTS exchange on the control similar to that of the disjoint flows. The throughput peaks at a
channel so as to reserve the access rights to transmit over certain bandwidth division and as the packet size increases,
the data channel. For broadcast, the node wishing to this optimum bandwidth division occurs at smaller control
broadcast a packet has to send an RTS on the control channel channel bandwidths. Regardless of the number of flows
before it can broadcast the packet on the data channel. The present in the system, the aggregate throughput the system
RTS is to inform all nodes in the network about the achieves also tends to converge as the control bandwidth
pending broadcast. When a node receives an RTS for a increases. while the general performance characteristics of both
broadcast, it will switch over to the data channel at the end scenario are the same, the absolute throughput achieved by
of the current transmission to receive the broadcast packet. this protocol in the AP flow scenario is significantly smaller
This protocol currently assumes that all nodes on a single than that of the disjoint flows and is always worse than IEEE
channel sense the same channel conditions, especially since the 802.11 DCF. This is due to the presence of the access point,
network is operating as a WLAN. While this assumption may which is the common receiver across all flows in the network.
hold for an isolated WLAN, the presence of other WLANs or Given that this protocol freezes a node’s back‐off counter if its
interference sources could result in different nodes sensing destination is known to be on the data channel, the contention‐
different channel conditions within the same wireless LAN. resolution stage is frozen during every data transmission cycle.
This impacts the transmissions on both the control and data This effectively sequentializes the contention‐resolution stage
channels. On the control channel, a node may not reply to an and the data transmission stage. Consequently, only one
RTS if it senses the channel to be busy, regardless of whether channel is in use at any instant. This bandwidth wastage
the packet was correctly received. This could again result in reduces the throughput that the system is able to achieve.
the sender’s CW exponentially growing unnecessarily. Nodes Furthermore, since the data channel bandwidth is always less
would also freeze their back‐off counters when they sense the than the total available bandwidth, this protocol will not be
58
able to match the performance of IEEE 802.11 DCF in the AP From the above discussion, it can be inferred that, for a
scenario. The this protocol throughput in this scenario can be random network topology, the benefit of this protocol is
improved by using two transceivers at the access point, which dependent on the number of flows and the packet size
will allow pipelining to proceed while it is receiving on the distribution as well as the bandwidth allocation between the
data channel. two channels. Despite the fact that the maximum throughput
Unlike the disjoint flows and the AP flows, random flows occurs at a certain bandwidth division, the decrease in the
are more realistic as they account for the possibility that a node aggregate throughput for a bandwidth division around the
could be both a sender as well as a receiver. This optimum is not very significant. This gradual degradation
interdependence will affect the aggregate throughput as implies that, given the packet and flow distributions, a certain
deafness becomes an important factor. While the performance bandwidth division can be chosen such that the system
curves for the random flows follow the same trends as those of throughput across all number of flows is close to the peak
the disjoint and AP flows, the throughput achieved by this throughput. Hence, the improvement of this protocol over
protocol for random flows lies within the range bounded by IEEE 802.11 is more sensitive to the number of flows in the
the throughput of the disjoint flows and that of the AP flows. network and to the packet size distribution than it is to the
As described earlier, the random flows were picked such that bandwidth division.
the each node in the network is a source of exactly one flow
The performance characteristics of the proposed protocol
and its destination is randomly selected from the remaining
in various network topologies via simulations are
nodes in the network. For a small number of flows, most of the
discussed. The simulations were done on a modified
flows become interdependent. Two flows in a network are
version of the ns2 network simulator from each data point is
independent if there are no common nodes between them;
an average of 10 runs. The total bit rate is set to 11 Mbps,
otherwise they are interdependent. Consequently, the
which is then divided between the two channels this
contention resolution and data transmission stages become
protocol requires. The physical header sizes were scaled
heavily sequentialized. The network becomes very similar to
appropriately so as to account for the 192‐microsec overhead
that of the AP flows and hence the performance of this protocol
incurred by transmitting the physical headers at the base rate
is worse than IEEE 802.11 DCF for a small number of flows.
of 1 Mbps. The carrier sense mechanism was modified to
As the number of flows increases, there are a larger
include one slot durations as the interval needed to
number of flows that are independent of each other. With a
determine whether or not the channel is idle. To maximize
greater number of independent flows, the impact of pipelining
the channel utilization, all flows adopt an aggressive
becomes more obvious. When one flow is on the data channel,
sending rate such that each always has backlogged
all the flows that are dependent on this flow will freeze their
packets to be sent. All the networks simulated are WLANs,
back off counters and not participate in the current round of
i.e., every node is within transmission range of every other
contention. Flows that are independent of the nodes currently
node in the network. The traffic sources are chosen to be
on the data channel proceed with the contention‐resolution on
constant bit rage (CBR) sources.
the control channel. There is a high likelihood that, by the end
Three different types of network topologies were
of the current data transmission cycle, one of the independent
simulated. The first type is the disjoint flow scenario,
flows would have won the contention‐resolution on the control
where each node is either the sender or the receiver in
channel and would be ready to switch to the data channel to
exactly one flow. The second type is the access point (AP) flow
start transmitting. Thus, the presence of independent flows
scenario, where one node is the receiver for all the flows
within a network causes the random flow scenario to behave
within the network. Finally, the random flow scenario is
like the disjoint flow scenario, thus resulting in an increase in
also simulated. Here, each node is the source for exactly one
the system throughput. This is evident from the sharp increase
flow and its destination node is randomly selected from the
in system throughput from a four‐flow scenario to an eight‐
rest of the nodes within the network.
flow scenario. As the number of flows continues to increase,
On the other hand, if the data transmission cycle is
the throughput gain that results from the presence of
significantly longer than the contention resolution cycle, it
independent flows within the network is slightly offset by an
implies that the bandwidth division is again suboptimal,
increase in the number of collisions. Despite this, the presence
causing the data to take a very long time to transmit. In such
of independent flows within a network still contributes to a
cases, while the utilization of the data channel is
significant performance gain. The performance of a network
maximized, the achievable throughput is limited by the
with 128 flows is always better than that with two or four
small bandwidth assigned to the data channel.
flows, regardless of the data size.
Decreasing the control channel bandwidth increases the
The performance of this protocol degrades below that of
throughput since the bottleneck on the data channel is
IEEE 802.11 faster for random flows than it does for disjoint
reduced. Thus, the peak throughput occurs at the
flows. This is indicative of the impact the interdependence and
bandwidth division that approximately balances the
deafness have on the overall system throughput. For a packet
contention resolution cycle with the data transmission
size of 750 bytes, the performance of DMCAC for an eight‐flow
cycle.
network becomes worse than that of IEEE 802.11 as the control
bandwidth becomes larger than 5 Mbps. In contrast, this
protocol performed better than IEEE 802.11 up to a control
bandwidth of 5.5 Mbps for an eight‐flow network in the
disjoint flow scenario. This implies that the performance gain is
more sensitive to the bandwidth division in the random flow
scenario than in the disjoint flow scenario.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151‐9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 59
Fig.
3: Throughput analysis of proposed method.
Fig. 5: Throughput analysis of proposed method.
As the packet size increases, the optimum bandwidth
division occurs at smaller control channel bandwidths, For a fixed packet size, as the number of flows in the
regardless of the number of flows present in the network. network increases, the optimal bandwidth division occurs
For data packets of 250 bytes, the optimum control channel at larger control bandwidth values. For example, given a
bit rate occurs in the range of 2.5 to 3.5 Mbps (i.e., packet size of 512 bytes, the peak throughput for 4 to 32 flows
approximately 23% to 32% of the total bit rate). For packet occurs around a control channel bandwidth of 1.5 Mbps,
size of 1000 bytes, this range shifts left to 0.5 to 1.5 Mbps (i.e., while the peak throughput for 64 and 128 flows occurs at a
5% to 14% of the total bit rate). This shift is due to the fact control bandwidth of 2.5 Mbps. This is because the number of
that as the packet size increases, the packet transmission time collisions on the control channel increases with the number of
also increases. As such, more bandwidth can be allocated to flows and, as a result, the average contention‐resolution
the data channel to reduce the absolute value of the data duration increases. Consequently, the control bandwidth
transmission cycle. With longer data transmission cycles, the has to be increased in order to maintain the balance
duration of the contention‐resolution cycles can be allowed between the contention‐resolution cycle and the data
to increase proportionately by decreasing the bandwidth transmission cycle.
allocated to the control channel. Thus, the equilibrium point The two‐flow scenario is an exception to the above
occurs at a smaller control channel bandwidths. trend as its peak throughput occurs at a larger bandwidth
The aggregate throughput curves tend to converge as the division than for the four‐flow scenario. For two flows,
control bandwidth increases. This is expected because, as a there is usually only one flow on the control channel. As a
larger fraction of the bandwidth is allocated to the control result, the expected idle period is 16 slots. As the number of
channel, the data transmission cycle becomes the flows increases, the probability that at least one flow has a
bottleneck in the system. Furthermore, as the packet size smaller back‐off value increases and thus the expected idle
increases, the duration of the data transmission cycle also period is also smaller. The reduction in the idle overhead
increases, which causes the bottleneck on the data channel reduces the contention‐resolution cycle, and consequently
to become more pronounced. Fig. 4, 5 and 6 shows the the optimal bandwidth division occurs at a smaller control
throughput analysis of proposed technique. channel bandwidth. But as the number of flows increases,
the larger number of collisions results in longer
contention‐resolution cycles and thus causes the optimal
bandwidth division to shift right.
The protocol masks the overhead introduced by the
contention resolution, particularly the idle duration that is
a result of the random back‐off procedure, with the ongoing
data transmission. But, as the control bandwidth increases,
the performance of this protocol eventually becomes worse
than IEEE 802.11 because the system throughput is restricted
by the small bandwidth allocated to the data channel. This
is seen from the graphs where this protocol performance
curves drop below those of IEEE 802.11. This performance
degradation becomes more pronounced as the packet size
increases. As the packet size increases, the performance of
this protocol becomes worse than IEEE 802.11 at
Fig. 4: Throughput analysis of proposed method. progressively lower control bandwidths. This effect is more
obvious for a smaller number of flows.
60
For example the performance of this protocol in a divides the available bandwidth between two channels,
network with four flows degrades below that of IEEE namely a control channel and a data channel. The contention‐
802.11 around a control bandwidth of 5.5 Mbps, given a resolution stage occurs over the control channel and involves
packet size of 512 bytes. When the packet size increases to the RTS/CTS exchange to win the access rights to transmit the
1000 bytes, beyond a control bandwidth of about 4.7 data packet. Upon winning the contention‐resolution, the
Mbps, IEEE 802.11 performs better than this protocol. But DATA/ACK exchange is done on the data channel. By
when there are 128 flows in the network, the performance of allowing the contention‐resolution for the next data packet to
this protocol only degrades below that of IEEE 802.11 at occur in parallel with the current data transmission, a
around 5.9 Mbps and only when the packet size is 1000 significant portion of the contention overhead is masked and
bytes. But this degradation occurs very far from the thus results in a throughput gain.
optimal bandwidth division. Thus, overall, this protocol Here simulated the protocol within an isolated WLAN
performs significantly better than IEEE 802.11 for disjoint while varying the packet size, bandwidth division, and
flows. number of flows. As expected, there is an optimum bandwidth
Finally, we consider the effect of details in visualization. division such that the duration of the contention‐resolution
We have developed nam as a generic tool for visualizing stage balances the duration of the data transmission stage, and
the output of network simulations. We find visualization a the aggregate throughput is max at this bandwidth division.
very important tool for protocol debugging, but there is need The control bandwidth at which, this optimum bandwidth
to control the amount of detail presented to the user. In this division occurs increases with the number of flows.
suggestion we examine ways we use visualization to Furthermore, as the bandwidth on the data channel decreases,
control details, and ways that visualization is helpful at the aggregate throughput that can be achieved tends to
selecting the right level of detail for wireless simulation. converge to a single value, regardless of the number of flows
Easy‐to‐use visualization alone provides a huge step and the packet size. These trends hold across all type of flow
providing a large amount of detailed information in a topologies. This technique results in a substantial performance
manageable fashion. improvement over IEEE 802.11 DCF for disjoint flows, but it
This resulted in MAC‐layer timeouts and anomalous performs significantly worse for the AP flow scenario.
behavior completely unrelated to the protocol we were The random flow scenario is a middle point between these
studying simply because we were out of an acceptable two scenarios and accounts for the common receiver problem
operating region. This status would have quickly and as well as the impact of deafness on the overall performance.
easily been determined from a protocol visualization, but Form the results, it is evident that the performance is more
was lost in the aggregate statistics we considered. Even sensitive to the distribution of packet size and to the number of
with visualizations, the detail can become overwhelming. flows than it is to the bandwidth division. The optimum
We are exploring two ways to control this detail in nam. bandwidth division for all number of flows occurs within a
First, we provide different kinds of visualization for range of 1Mbps of each other. Also, a slight deviation from the
different kinds of wireless communication. Second, we optimum bandwidth allocation does not significantly degrade
allow the user to control the level of detail nam presents. the throughput of the system. Hence, if the packet size
Nam has two ways to visualize wireless distribution is known, a fixed bandwidth division can be
communications. First, we can visualize packet flow as selected such that the system throughput across all number of
rectangles that are animated and move directly from the flows is relatively close to the peak value.
source to destination (the lines from node 1 to nodes 2 and 3 The next step would be to attempt to extend the pipelining
in Figure). This representation has evolved from namʹs use protocol to the multi‐hop scenario and evaluate the actual
to visualize wired point‐to‐point networks where packets impact differing channel conditions have on the aggregate
flow on links. This approach clearly identifies the sender and throughput. Another possible direction for future work would
receiver of the packet, the direction of packet flow, and the be to outfit each node with two transceivers. The evaluation in
time of transmission and receipt. However, this the thesis was done under the assumption that each node only
visualization does not easily adapt to support broadcast has one transceiver and is only able to send or receive on one
traffic. Representing a broadcast packet as multiple channel at any point in time. As such, dependent flows become
rectangles visually suggests multiple packets. This sequential zed since contention‐resolution cannot proceed
approach also does not easily show when concurrent while the common node is on the data channel. With two
transmissions from different nodes interfere with each transceivers, nodes will be able to send and receive on both
other. An alternate visualization approach is to show channels simultaneously. Consequently, a node will be able to
wireless packets as expanding circles (the circles in participate in the contention‐resolution process on the control
Figure). This clearly shows the packet source and channel while it is involved in the current transmission on the
interference with other packets, but it does not show data channel.
destinations.
5 REFERENCES
4 CONCLUSION AND FUTURE WORK [1] Xue Yang and Nitin H. Vaidya, ʺDSCR: A More Stable Wireless
This paper describes and evaluates the MAC protocol with MAC Protocol,ʺ Technical Report, Coordinated Science
many channels that incorporates and explicit pipelining Laboratory, University of Illinois at Urbana‐Champaign, Dec.
structure. This technique adapts the IEEE 802.11 DCF protocol 2002
by splitting the transmission dialog into a contention‐ [2] Xue Yang and Nitin H. Vaidya, ʺPipelined Packet Scheduling in
resolution stage and a packet transmission stage. The protocol Wireless LANs,ʺ Technical Report, Coordinated Science
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 7, JULY 2010, ISSN 2151‐9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 61
Laboratory, University of Illinois at Urbana‐Champaign, Aug. on. He has organized several AICTE sponsored National seminar/
2002. workshops.
[3] Xue Yang and Nitin H. Vaidya, ʺDSCR ‐ A Wireless MAC
Protocol Using Implicit Pipelining,ʺ ACM International
Symposium on Mobile Ad Hoc Networking and Computing
(MobiHoc) Poster Session, June 2003.
[4] X. Yang and N.H. Vaidya, ʺExplicit and Implicit Pipelining for
Wireless Medium Access Control (Invited Paper),ʺ Proc. IEEE
Semiann. Vehicular Technology Conf., Fall 2003.
[5] Yang Xiao, Hui Chen, and Mohsen Guizani, ʺPerformance
Evaluation of Pipeline Paging under Paging Delay Constraint
for Wireless Systems,ʺ IEEE Transactions On Mobile
Computing, Vol. 5, No. 1, pp.64‐76, January 2006
[6] Xue Yang, Member and Nitin H. Vaidya, ʺA Wireless MAC
Protocol Using Implicit Pipelining,ʺ IEEE Transactions On
Mobile Computing, Vol. 5, No. 3, March 2006
[7 ] Xue Ya ng, Ni ti n H . Vaidya a nd Priya Ravichandran, ʺSplit‐
Channel Pipelined Packet Scheduli ng for Wireless
Networks,ʺ IEEE Transactions On Mobile Computing, Vol. 5,
No. 3, Pp. 240‐25 7, March 2006.
[8] Giuseppe Bianchi, Luigi Fratta and Matteo Oliveri,
ʺPerformance Evalution and Enhancement of the CSMA/CA MAC
Protocol for 802.11 Wireless LANs,ʺ Politecnico di Milano, Dip.
Elettronica.
[9] Federico Calì, Marco Conti, and Enrico Gregori, ʺIEEE 802.11
Protocol: Design and Performance Evaluation of an Adaptive
Backoff Mechanism,ʺ IEEE Journal on Selected Areas in
Communications, Vol. 18, No. 9, pp. 1774‐1786, September 2000.
[10] Brian P. Crow, Indra Widjaja, Jeong Geun Kim and Prescott
T.Sakai, ʺIEEE 802.11 Wireless Local Area Networks,ʺ IEEE
Communications Magazine, pp. 116‐126, September 1997.
[11] Ye Yan, DongSun Park, ʺDesign of MAC Protocols with Less
Collision Fast Resolution for WLAN,ʺ IEEE, pp.571‐573, 2004.
V. Kumar has completed his Postgraduate degree in Computer
Science and Engineering from Dr. M.G.R Engineering College
(Affiliated to Madras University, Chennai), Chennai in 2003 and now
he is doing his research in Computer Networks. Presently he is serving
as a Professor in the Department of Computer Science and
Engineering at Knowledge Institute of Technology, Salem,
TamilNadu, India. He has around 4 years of Industrial and 9
years of Teaching Experience. He has attended and presented many
papers in various National and International Conferences. He is a Life
Member of ISTE.