Wireless–Optical Broadband Access Network

Document Sample
Wireless–Optical Broadband Access Network Powered By Docstoc
					                                      CHAPTER - 1

                                   INTRODUCTION

A Wireless–Optical Broadband Access Network (WOBAN) has a Wireless Mesh Network
(WMN) to connect to end users and an optical backhaul network to carry the aggregated traffic
collected over aWMN . In this way, a WOBAN can achieve cost effectivedeployment of a WMN
while having higherperformance due to the optical backhaul network. Atthe back end of the
network, composed of a Passive Optical Network (PON) , an Optical Line Terminal
(OLT)resides in the Central Office (CO) and is connected viaoptical fiber to multiple Optical
Network Units (ONUs).At the front end, a set of wireless nodes (routers)forms a WMN. End
users, both mobile and stationary,connect to the network through these wireless nodes,whose
locations are generally fixed. A selected set ofthese nodes, called gateways, are connected to the
opticalpart of the network. Usually, gateways are attachedto one of the ONUs . Figure 1 shows
the architectureof a WOBAN




           .


 An end user sends its data packets to a nearby wireless node of the WOBAN for upstream
communication. These packets travel through the WMN, possibly over multiple hops, and reach


                                                  1
the OLT via the gateways. (From the OLT, the packets can be routed to the rest of the Internet.)
Similarly, for downstream communication, data packets travel from the OLT through the WMN
to the end users. Because a packet may need to travel several hops through the WMN, delay in
the access network can become significant. Moreover, some links in the WMN may be over
utilized and others may be underutilized under inefficient routing. If the transmission capacity of
a wireless node is not properly distributed among its transmission links, the link delays may
become high. Thus, capacity assignment and packet delay are important factors to consider for
communication between wireless nodes. Moreover, because ONUs (collocated with the
gateways) can have different load, links from these ONUs to respective OLTs can have different
delays. Thus, selecting a gateway with lightly loaded ONUs can reduce the delay. Hence, routing
in the WMN of a WOBAN should incorporatethe load and delay on the PON links as well.
Note that routing in the WMN part of a WOBAN has differences from routing in a general
WMN since traffic from a wireless node to the “rest of the Internet” can be directed through any
gateway. Thus this “anycast” property can be exploited for routing traffic from the wireless nodes
in a WOBAN, thereby opening up new opportunities for improved routing and load balancing
(unlike the case in a general WMN where the traffic is unicast). By considering the delay in the
optical part of the WOBAN, we can achieve stronger optical–wireless integration.
Capacity- and Delay-Aware Routing (CaDAR) algorithm that properly routes packets in the
WMN to reduce the network-wide average packet delay in the WMN and PON using optimal
capacity assignment on the links and using delay-aware routing.
          An improved algorithm, called CaDAR, which exploits the work on Capacity
Assignment (CA) and Flow Assignment (FA), which were originally designed for general packet
networks using optimal capacity assignment and flow deviation on links to minimize delay. The
combined problem of Capacity and Flow Assignment (CFA), however, has an optimal solution
for a specific input flow and can be called local optimal.
          Our objective in this work is to design and investigate the properties of an efficient
routing scheme based on capacity assignment and delay on the links of a WOBAN. Our
contributions are (i) proposing a routing scheme, CaDAR, with capacity assignment and delay
awareness and (ii) comparative performance analysis of CaDAR with the DARA and CFA.Our




                                                   2
study shows that CaDAR supports almost the same load on the network as CFA while having
lower delay than the DARA and a lower average number of wireless hops than CFA.




                                              3
                                 CHAPTER - 2
   CAPACITY- AND DELAY-AWARE ROUTING

CaDAR is a routing algorithm for the WMN of a WOBAN that minimizes the network delay by
assigning radio capacities and link weights based on link states. Two wireless nodes have a link
between them if their distance is less than their respective transmission range. Each node has one
radio, so a node’s limited radio capacity needs to be distributed among its outgoing links. Each
node advertises the states of all of its outgoing links using Link-State Advertisement (LSA).
Based on the LSA, capacity and delay are assigned to links and the shortest paths between
wireless nodes and gateways are calculated.
2.1 Capacity Assignment in CaDAR


Time Division Multiple Access (TDMA) has been a popular multiplexing scheme in
telecommunication networks, so it is a good choice for the WMN in a WOBAN for
communication between the wireless nodes. The number of time slots in a TDMA frame depends
on the number of links induced from a wireless node. Consider the example in Fig. 2. There are
three nodes originated from node u in Fig. 2, so the radio at node u operates on three time slots
(which need not be equal). Similarly, node w has two time slots.




Assignment of different time-slot durations for differentinks induced from a node translates to
assignmentof different capacity on each link. Noting thecurrent flow on each wireless link from



                                                  4
the LSAs, we determine the fraction of a node’s radio capacity thatshould be assigned to each
link originated by itself such that the network-wide average delay is minimized. The sum of
capacities on each outgoing link from a wireless node must be equal to the capacity of the radio at
the node. Capacity assignment is optimal for the flow given by the LSA. If a link does not carry
any flow on it, the capacity assignment of CaDAR automatically treats that link as nonexistent
and the time-slot duration for that link is assigned as zero forthat LSA period. Note that the time
slots and their durations can be assigned dynamically.


Figure 2 shows an example of capacity assignmentwhere node u has three neighbors v, w, and x.
In general,the loads on node u’s transmission links, λuv, λuw,and λux, could be different. To
minimize delay, the radiocapacity at node u should be distributed properlyamong the links (u,v),
(u,w), and (u,x) based on theirrespective flows. This can be achieved through assigningdifferent
time-slot durations for the different linksbased on their flows. Hence, capacities on these
links,Cuv,Cuw, and Cux, need to be assigned based on λuv, λuw, and λux. Similarly, for the radio
at node w, Cwuand Cwxdepend on λwuand λwx. Note that Cuwis assignedfrom the radio at node
u, and Cwuis assignedfrom the radio at node w. Moreover, because upstream and downstream
traffic flows could be different, λuw≠ λwu. Thus, links (u,w) and (w,u) are treated differently.


As described in Section I, in a WOBAN, a wireless node communicates with the “rest of the
Internet” through the OLT. Thus, when the wireless node sends a packet toward the OLT, the
packet can take any route to any of the gateways, and the gateway delivers the packet to the OLT.
This anycast nature of the routing, illustrated in Fig. 3, is exploited in our work.




                                                    5
Because of anycast routing, a packet travels from a wireless node s to any gateway gє G, where G
is the set of gateways. For any source–destination pair (s,g), let the traffic between them be γsg. If
the number of nodes in the WMN of a WOBAN is N and the total traffic is γ, then




The load on a link is defined as the amount of traffic on a link. In Fig. 2, if γvwand γuwgo
through link (u,w), then λuw=γvw+γuw. Let ω(N) be the set of N wireless nodes in the WOBAN.
Then, assuming independentarrivals, system delay, T, can be approximatelystated as follows :




where packet lengths are exponentially distributed and 1/µ is the average packet size. If each term
is convex in Eq. (2), their sum is also convex. For µCuv −λuv>0, first and second derivatives of
Eq. (2) are nondecreasing and nonnegative, respectively. Because the flow on any link (u,v), λuv,
cannot be more than its capacity, Cuv, Eq. (2) is convex.
As depicted in Fig. 2, capacities of links (u,v), (u,w), and (u,x) share the radio capacity at wireless
node u, ζu. Thus, the total capacity of all links induced from a node is equal to the capacity of the
radio at the node. If η(u) is the neighbor set of node u, then Av є η(u), and we have




We want to assign Cuvsuch that T is minimized for a given γ [Eq. (1)] where the sum of Cuvhas
to be equal to ζu. From Eqs. (2) and (3), we have the Jacobian as follows:




whereluis the Lagrangian multiplier. We can solve Eq. (4) for any wireless node u in a WOBAN
using




                                                    6
From Eqs. (4) and (5), we obtain




We obtain λuvfrom LSA messages. We adjust the time-slot duration τuvfor link (u,v) based on
Cuv. If the TDMA time frameduration for node u is u, then




   2.2 Delay Awareness in CaDAR: Wireless Part


The delay from the end user to the wireless node is unavoidable and relatively small compared
with the delay incurred in the mesh because a packet may travel through




several hops in the WMN before reaching a gateway. Packet delay in the WMN of a WOBAN
has four parts:
1) Transmission delay: As described earlier, each link can have different capacity. Transmission
delay on a link is given by 1/µCuv, where 1/µ is the average packet size.
2) Slot-synchronization delay: Slot-synchronization delay is associated with TDMAbased
operation of a wireless channel. Each router needs to send a packet to its neighbor in its
designated time slot. This delay occurs because an arriving packet needs to synchronize to the
time slots for communication with the neighbor of the wireless node. Average slot-
synchronization delay in a TDMA system is 1/2µCuv.



                                                  7
3) Queuing delay: Queuing delay depends on the service rate and packet arrival rate at a wireless
node. Higher capacity and a lower arrival rate leads to lower delay. The queueing delay
accumulates as a packet travels through several wireless nodes before reaching the destination
node. Delay in a queue can be expressed as λuv/µCuv(µCuv−λuv) if the arrivals are independent.
4) Propagation delay: In a WMN, the routers are close to one another; hence the propagation
delay is negligible. We approximate the packet arrivals at the wireless nodes to be independent.
Noting that the propagation delay on a wireless link is negligibly small compared with other
delay components for typical settings, the average packet transfer delay, or link delay, on any link
(u,v), depends on the transmission delay from node u to node v, as well as slot-synchronization
delay (for TDMA) and queueing delay at node u. Thus, delay on any wireless link (u,v) is given
by




The total delay for packet transmission between two nodes equals the sum of the delays on each
link in the path between the two nodes. We assign the link delays as weights of corresponding
links and compute the shortest paths to the gateways from each node and vice versa. This gives
the shortest-delay path for each node to the gateways to incorporate delay awareness of the WMN
part of a WOBAN in Ca- DAR. Then, the anycast routing to the OLT carries a packet to the
gateway that has the shortest-delay path from the wireless node.
            As shown in Fig. 2, if the shortest-delay path from node v to w is (v,u,x,w), then the
wireless path delay from v to w can be calculated from Eq. (8) as




                                                   8
2.3. Delay Awareness In CaDAR :Optical Part


In a WOBAN, gateways are collocated with ONUs (Fig. 1). These ONUs may be connected to
some other entities, such as another wired or wireless network such as a LAN or a WiMax node.
Thus, an ONU receives traffic not only from the gateway of the WMN of a WOBAN but also
from other networking entities. Thus, load on ONUs of the optical part of a WOBAN depends on
the network design and traffic from other network entities.


In Fig. 4, the WOBAN is shown to have two gateways. Each gateway is connected to an ONU,
which is connected to several other network entities as well.The OLT is serving several other
ONUs as well. Loadon ONU 1 depends on the traffic from the storage areanetwork (SAN),
WiMax base station, and Gateway 1 of the WOBAN, which can be different from the load on
ONU 2 that depends on the traffic from the LAN and Gateway 2. Thus, delay from ONU 1 to the
OLT and from ONU 2 to the OLT can be different, and selecting a proper gateway through which
to route traffic can become significant.




         CaDAR works with Interleaved Polling with Adaptive Cycle Time (IPACT) for
communication between ONUs and the OLT. CaDAR is used to find the capacity- and delay-
aware route for a packet from the OLT to the ONU, and IPACT delivers the packet to the
respective ONU. IPACT can perform dynamic bandwidth distribution and can use a single



                                                  9
downstream and a single upstream wavelength with an ability to provision a fractional
wavelength capacity to each user. There are several service disciplines for IPACT. Among them,
the constant credit service scheme has lower delay than other schemes. In this scheme, an ONU
can send a fixed (constant) amount of additional traffic beyond its allocated timeslot size. As the
WMN part of a WOBAN continuously collects traffic and sends them to the gateways, this
scheme allows the accumulated traffic to be sent with the added time slot. Thus, for our work, we
assume that ONUs send data to OLTs using IPACT with constant credit service. We see the
impact of load on delay over the optical link between the ONU and OLT. Hence, if the load on an
ONU is known, the delay on the optical link can be estimated. IPACT uses broadcast for
downstream and point-to-point for upstream communication. Thus, we consider both the aspects
for the two directions of traffic flows.
1. Downstream Communication: The broadcast nature of downstream traffic lets each ONU
receive packets destined to every other ONU. Hence, each gateway receives packets destined for
all the gateways in the WOBAN, keeps the packets destined to itself, and discards the rest, as
shown in Fig. 5. From the gateway to the wireless node, the traffic is unicast. The OLT needs to
select the gateway such that the packets travel the shortest-delay path from the OLT to the
destined wireless node. Gateways send LSAs to the OLT so that it selects the correct gateway.
The OLT considers both optical and wireless delay for gateway selection. The delay on the
downstream optical link includes the transmission delay and the propagation delay. If the
capacity on all the optical links between the ONU and OLT are the same, then the transmission
delay is going to be the same for any particular packet. Thus, transmission delay is not considered
for gateway selection.




                                                 10
              Let the distance from the OLT to an ONU collocated with a Gateway g be Lg, the
propagation delay on the optical link for unit distance be dp, and the shortest-delay path from
Gateway g to a node u be Pgu. To select the gateway, the OLT calculates the downstream delay
as follows:




2. Upstream Communication: CaDAR considers the load on the collocated ONU of the
gateway while routing packets from users to a gateway. IPACT is used for point-to-point
communication from the ONU to the OLT. Even if there is a shortest-delay path to a gateway,
that gateway can be heavily loaded and can have high delay in the optical part. Thus, while
forming the shortest-delay route to the OLT, calculating the shortest-delay path in WMN [Eq.
(9)] is not sufficient. Instead, the wireless node sends the packet to the OLT, and anycast routing
brings the packet to one of the gateways that gives the shortest delay from the wireless node to
the OLT. To calculate the delay on the optical part, a gateway obtains the load information from
its collocated ONU and estimates the delay on the optical part. Then, it broadcasts this
information with its LSA, and this information is propagated through the WMN.
IPACT grants each ONU a slot to send its upstream data. Thus, the transmission goes through
ON/OFF periods. The delay on the upstream optical link involves waiting time, transmission
delayand propagation delay. Waiting time depends on how the on/off periods are distributed and
on the load of the network. Let the set of all ONUs served by the OLT be O, the distance between
any ONU o and the OLT be Lo, load on any ONU o be γo, transmission capacity on the links be
R, and the shortest-delay path from a wireless node u to a Gateway g be Pug. An ONU needs to
wait while the OLT serves all other ONUs. Assuming each ONU has a nonzero load, the
minimum waiting time for any ONU o is




For alternating Pareto-distributed on/off periods , with shape parameter k, the mean waiting time
for any ONU o is


                                                 11
           At the ingress wireless node, to select the upstream route in the WOBAN, the shortest-
delay paths are computed through each gateway as




wheredgWTis the mean waiting time for the ONU collocatedwith Gateway g. A packet from the
ingress router u goes to the Gateway g that minimizes Eq.(13) following the shortest-delay path
Pug. When thepacket reaches the gateway, IPACT sends the packet to the OLT. Figure 3
illustrates the upstream routingin CaDAR




2.4 Link-State Advertisement


CaDAR operates on current flows on different wireless links in a WOBAN. Each node sends out
periodic LSA messages to its neighbors, notifying the states of its link. We assign the capacities
and the weights on the links for routing.
1. Weighted Moving Average: If the link states vary frequently, it means that the traffic is
changing rapidly. In such a case, it is not advisable to change the capacity and the link weight
based on the flow obtained from LSAs; instead, we maintain an estimated flow value using the
weighted moving average (WMA). If a router receives an LSA, say the kth one from its
neighbors, and obtains the flow on link (u,v) as λkuv, we can calculate the estimated flow on
thatlink based on the WMA, eλkuv, as




                                                 12
Where α is the decaying index of the WMA and S is the number of samples used per prediction.
We assign Cuvk based on eλuvk and calculate Wuvk with eλuv k instead of λuvk


2. Overhead: It is shown that for frequent LSA periods, the bandwidth consumptions due to
LSAs in a WOBAN are a small fraction of the total bandwidth of links. This fraction becomes
even smaller when higher-capacity radio (e.g., 54 Mbps instead of 11 Mbps) is used, as is done in
our performance is studied, because the size of the LSA messages remains the same. Hence, we
see that the LSA overhead for CaDAR is low.
2.5 Impact Of Traffic Change
Each wireless node usually follows a particular trend. For example, a node in downtown will be
highly loaded in daytime during a weekday and lightly loaded at night. WMA combines the
overall trend of the load in the WOBAN with the instantaneous load to determine the weighted
load and optimizes the wireless capacity based on this weighted load. This mitigates the sudden
burstiness of traffic and prevents drastic changes in routes and avoids further changes to load
conditions.
CaDAR incorporates anycast routing (Fig. 3) for upstream and a combination of broadcast and
unicast routing (Fig. 5) for downstream in the WOBAN. It accommodates traffic variations using
LSAs. CaDAR is described in Algorithm 1.




                                                13
                                        CHAPTER - 3
                          ILLUSTRATIVE EXAMPLES

We apply CaDAR on the 43-node wireless network with three gateways in the Wildhorse area of
Davis, CA, as shown in Fig. 6. The topology chosen for performance evaluation consists of actual
locations of wireless routers in the Wildhorse neighborhood of Davis, CA. Though these routers
are not connected to form a wireless mesh network today, we have used their locations to obtain a
real distribution of wireless routers over a large area. Three gateways are collocated with three
ONUs served by an OLT at the central office, which also serves several other ONUs performing
other operations. Each node is equipped with one radio with a capacity of 54 Mbps, as in IEEE
802.11 g. We use k=1.4 , number of ONUs =16, and downstream and upstream data rates of PON
as 1000 Mbps and 100 Mbps, respectively, in our numerical examples. We considered each ONU
to be 10 km from the OLT, hence giving 50 µs of propagation delay on the optical links.




                                                14
15
Algorithm 1CaDAR Algorithm
      Link-State Advertisement(LSA)
       1. .For each link (u,v) from node u , advertise periodically the capacity (Cuv), flow
             (λuv),and time stamp to wireless nodes.
       2. If node u is a gateway , then
       3. Advertise the delay on shortest-delay path to each node v ≠ u in topology to the OLT
       4. Advertise the optical-link delay from the collocated ONU to the OLT to the wireless
             nodes
       5. end if
      Link-Weight Assignment
       1. Receive any LSA , say the kth one , from neighbors
       2. Determine flow kthλuv on all links (u , v) for the kth LSA
       3. Calculate ktheλuv on all links (u , v) using kthλuv and k-1 theλuv
       4. Assign capacity k thCuv based on ktheλuv
       5. Assign weightWkuv=duv=1/µCKUV+1/2CKUV+1/µCKUV-eλkuv
      Path Computation


       1.    If traffic is downstream,then
       2.    For any packet destined to node u, calculate dDNg=Lgdp+Σ(u,v)єPguduv for all gateways
       3.    Send the packet to gateways g that minimizes dDNg
       4.    Calculate the shortest-delay path from gateway g to the node u based on the link
             weights   Wkuv
       5.    else if Traffic is upstream, then
       6.    For any packet originating from node u, calculate the shortest – delay path from node
             u to each gateway g base on the link weights Wkuv
       7.    Calculate dUPg=Lgdp+dgWT+Σ(U,V)єPugduvfor all gateways
       8.    Send the packet to Gateway g that minimizes dUPg
       9.    Gateway g sends the packet to the OLT through the collocated ONU
       10.   end if



                                                    16
We compare the performance of CaDAR with the DARA and a static algorithm, CFA, which
gives an optimal solution for a specific input flow. The results show that the DARA outperforms
routing approaches based on the shortest path, minimum hop, and throughput; hence,
performance improvement of CaDAR over the DARA is a very significant result.
            Each node has a load, which is the rate of traffic from the optical network of a
WOBAN to that wireless node. The upstream load is two-thirds the downstream load. Thus, in
Fig. 7, the load at each node indicates 60% downstream and 40% upstream traffic.
            CaDAR performs integrated routing on both the wireless and optical parts of a
WOBAN, whereas the DARA operates only on the wireless part. CFA is not proposed for hybrid
networks. Thus, first, we show the performance analysis of CaDAR, the DARA, and CFA
considering equal load on each ONU where each ONU carries only the traffic from its collocated
gateway, so while routing, the impact of optical links can be ignored.


Figure 7 shows the system delay, which is average network-wide packet delay, for the DARA,
CaDAR, and CFA for different traffic loads at the wireless nodes. We observe that CaDAR can
support nearlythree times the maximum load that the DARA can handle while still maintaining
low system delay. We also see that CaDAR performs almost similar to the local optimum (CFA)
for a load at each node up to 3 Mbps and follows very closely for higher loads till about 3.5
Mbps.




                                                  17
CFA outperforms CaDAR at higher loads because CFA readjusts the flows and capacities to
obtain better throughput and delay on links than CaDAR. CFA does flow distribution on the links
iteratively while adjusting capacities on the link unless the local optimum is reached. (This
explains CFA’s delay discontinuity around a load of 3.6 Mbps in Fig. 7.) On the other hand,
CaDAR performs capacity assignment and does fixed routing (assigning flows to links) based on
the current LSA.
Figure 8 shows individual path delays among thethree schemes. For each scheme, we show the
delay onthe longest-delay path in Fig. 6. We find that, whileDARA can carry up to 1.75 Mbps on
a single path, Ca-DAR can support twice the load. Also, the delay on thelongest path is higher
than the average delay for allthree schemes. Moreover, as expected, even thoughCaDAR can
carry a load of nearly 3.5 Mbps on average,it can carry about 3.2 Mbps for the longest path.
Figure 9 shows the average number of hops for thethree algorithms. As a WMN is the major
contributorto contention and delay in a WOBAN, fewer hops in aWMN translates to lower delay.
We observe that Ca-DAR and the DARA have an overlapping averagenumber hops as both of
them perform delay-awarerouting. On the other hand, CFA deviates flows on thelinks to decrease




                                                 18
the delay, which in turn may increasethe average number of hops more than CaDARin the
network.




Figure 10 shows the load balancing of CaDAR, theDARA, and CFA. We plot the traffic
difference, whichis the difference between the maximum and the minimumpacket intensities for
links in the WMN. Thesmaller the difference, the better the load balancingwill


                                                 19
be (or the less the link congestion will be) and viceversa. Though CaDAR and the DARA
perform delayawarerouting, CaDAR has lower link congestion. Fora higher load, CaDAR has
lower flow and capacity onthe links than CFA, which is reflected in Fig. 10. Wecan conclude that
CaDAR performs better load balancing,irrespective of the load.
       Next, we investigate the impact of delay on the optical part of a WOBAN. We allocate
different loads on the collocated ONUs of the three gateways in Fig. 6. We make the ONU
collocated with one of the gateways 80% loaded and the ONU collocated with the other two 20%
loaded. This causes higher delay on the optical link from one ONU to the OLT. As a result,
Algorithm 1 diverts wireless traffic from the gateway collocated with the overloaded ONU
toward othergateways to maintain shortest-delay paths. Hence,the load on ONUs influences the
traffic of the WMN ofthe WOBAN and in turn influences the load on thegateways.


Figure 11 shows the system delay for CaDAR when the ONUs collocated with the gateways are
both equally (Fig. 7) and unequally loaded. When the load on a particular ONU is increased,
more traffic is diverted toward the other gateways and their collocated ONUs, and packets travel
longer paths, leading to higher system delay. We observe that, if the load on an ONU increases,
the system delay of the network increases and the maximum traffic supported by a node
decreases. Thus, CaDAR incorporates the optical delay information and performs delay-aware
routing over the entire WOBAN.

    Load Distribution on
           ONU                   Gateway 1              Gateway 2              Gateway 3
           Equal                   35%                     35%                    30%
      ONU 1 overload              12.5%                    40%                    47.5%
      ONU 2 overload              37.5%                   22.5%                   40%
      ONU 3 overload              42.5%                   37.5%                   20%


   TABLE 1
   Fraction of paths through a gateway for different ONU loads



                                                20
Table I shows the fraction of paths between thewireless nodes and the OLT through any
particulargateway. As traffic between each wireless node andthe OLT travels through one of the
gateways, we observe from Table I how the traffic isdistributed in the WMN of the WOBAN
when anyONU is overloaded. For a moderate load of 1 Mbps ateach wireless node, supported by
all the cases shownin Fig. 11, we see that, if the ONUs have equal loads,35%, 35%, and 30% of
paths between the wirelessnodes and the OLT are established through Gateways1, 2, and 3,
respectively. On the other hand, whenONU 1 is overloaded, only 12.5% of paths between
thewireless nodes and the OLT are through its collocatedGateway 1. When ONUs 2 and 3 are
overloaded,22.5% and 20% of paths are established through Gateways2 and 3, respectively,
which are almost doublethe number of paths through Gateway 1 when its collocatedONU 1 is
overloaded. This is because Gateways2 and 3 are connected to a number of wirelessnodes via a
single hop (Fig. 6); as a result, theshortest-delay paths for these wireless nodes are createdthrough
Gateways 2 and 3 even when their collocatedONUs (2 and 3) are overloaded.

Figure 12 shows the average number of hops between wireless nodes and gateways for CaDAR
with equal and unequal loads at the ONUs. We observe that, when the ONUs have equal load,
traffic from wireless nodes tends to go toward the closest gateway using fewer hops because this
usually gives the shortest-delay path. On the other hand, when the


loads on the ONUs become different, the average number of hops in a path increases because
packets travel to relatively distant gateways (Fig. 6). Similar to the observations from Table I,
when ONU 1 is overloaded, the packets tend to travel more hops than when ONUs 2 and 3 are
overloaded




                                                  21
                                            TABLE 2
Overload        Load at each F0                 F(0-1)           F(1-0)          F>1
ONU             Node
ONU 1           1 Mbps          69.84%          18.65%           9.13%           2.38%
ONU 2           1Mbps           67.745          23.79%           8.06%           0.81%
ONU 3           1Mbps           68.55%          22.58%           8.06%           0.81%
ONU 3           1.5Mbps         68.95%          19.76%           8.47%           2.82%
ONU 3           2Mbps           68.95%          9.68%            16.13%          5.24%


Distribution(%) of Flow on Wireless Links for CaDAR


The first three rows of Table 2 show the percentage distribution of load on different wireless links
for Ca- DAR with a moderate load of 1 Mbps at each node. Here, we observe that most of the
wireless links do not carry any traffic. The general trend is that, when the loads on the ONUs are
different, more traffic is diverted toward the ONU with lower load. As a result, all the links


                                                  22
leading to their collocated gateways become heavily loaded. In the last three rows of Table 2, we
see that for a particular overloaded ONU (ONU 3), with increasing load, the number of links
carrying higher flow is increased. We observe that, when the load at each wireless node increases
from 1 to 2 Mbps, flows of 2 Mbps or higher on a link increase from less than 1% to more than
5%. As load increases, some links, especially those closer to the gateways, tend to carry higher
flows and become the bottleneck for the WMN of the WOBAN.


CaDAR utilizes the architecture of the WOBAN that enables a faster convergence of the
algorithm because the LSA from one part of the wireless front end can travel over the optical
backhaul to reach the other parts of the network instead of taking all the wireless hops. For
example, in the topology in Fig. 6, it takes 12 wireless hops to travel from node 12 to node 29.


But in a WOBAN, it takes just four wireless hops and the rest is traveled over the optical
backhaul. It is notable that any wireless node can reach the optical backhaul in a maximum of
three hops in the topology in Fig. 6.


Figure 13 shows the delay incurred by the LSA from node 6 to node 38. We select these two
nodes because they have a maximum of six wireless hops between them. We observe that for
different loads, the delay for the LSA remains below 0.45 ms even for high load. In this delay
calculation, we considered the links loaded with regular data traffic and added the LSA overhead
with that.




                                                  23
This shows that even for this worst-case scenario, the LSA information reaches from one node to
another quickly. Because CaDAR uses WMA to compute the path, these fast-traveling LSAs
along with the gradual changes of trafficallow a stable performance from CaDAR.




                                               24
                                        CHAPTER - 4

                                       CONCLUSION

A WOBAN tries to combine the cost-effectiveness of a WMN and the high capacity of an optical
network. But a WMN introduces delay and capacity bottleneck. To gain the desirable
performance from a WOBAN, we have proposed an efficient routing scheme for a WOBAN
called Capacity- And Delay-Aware Routing (CaDAR), which performs capacity- and delay-
aware routing to support higher loads in the network and to minimize packet delay. CaDAR
optimally distributes the radio capacity of a single-radio wireless node among its outgoing links
and performs delay-aware routing in a WOBAN. It incorporates delay awareness of both the
wireless and optical parts of a WOBAN to find the shortest-delay path through the entire
WOBAN. Our illustrative examples show that CaDAR performs very close to the local optimal
solution while minimizing the average number of hops in the WMN of a WOBAN.




                                                25
                                   CHAPTER – 5

                                  REFERENCES

1. S. Sarkar, S. Dixit, and B. Mukherjee, “Hybrid wireless-optical broadband access network
   (WOBAN): a review of relevant challenges,” J. Lightwave Technol., vol. 25, no. 11, pp.
   3329– 3340, Nov. 2007.
2. Effenberger, D. Clearly, O. Haran, G. Kramer, R. Li, M. Oron, and T. Pfeiffer, “An
   introduction to PON technologies,” IEEE Commun. Mag., vol. 45, no. 3, pp. S17–S25,
   March 2007.
3. G. Narlikar, G. Wilfong, and L. Zhang, “Designing multihop wireless backhaul networks
   with delay guarantees,” in IEEEINFOCOM, Barcelona, Spain, April 2006, pp. 1–12.
4. S. Lee, G. Narlikar, M. Pal, G. Wilfong, and L. Zhang, “Admission control for multihop
   wireless backhaul networks with QoS support,” in IEEE Wireless Communications and
   NetworkingConf. (WCNC), Las Vegas, NV, April 2006, pp. 92–97.
5. T. Liu and W. Liao, “Capacity-aware routing with multichannel multi-rate wireless mesh
   networks,” in IEEE ICC, Istanbul, Turkey, June 2006.




                                            26