Sie sind auf Seite 1von 6

Multipath Routing in Wireless Mesh Networks

Nagesh S. Nandiraju, Deepti S. Nandiraju, and Dharma P. Agrawal


Center for Distributed and Mobile Computing, Dept. of ECECS, University of Cincinnati - Cincinnati, OH 45221
(nandirns, nandirds, dpa)@ececs.uc.edu

Abstract in design with the only exception that a mesh


Wireless Mesh Networks are envisioned to support the point is connected to the wired network and thus
wired backbone with a wireless backbone for providing also called as Internet gateway. Figure 1 shows a
internet connectivity to residential areas and offices. sample mesh network in a university.
Routing protocols designed for Mobile Adhoc Networks
(MANETs) primarily concentrate on finding a single
best possible route to any destination out of the various
paths available. However in wireless mesh networks,
traffic is primarily routed either towards the Internet
Gateways (IGWs) or from the IGWs to the Access Points
(APs). Thus, if multiple APs choose the best throughput
path towards a gateway, the traffic loads on certain
paths and mesh routers increases tremendously thereby
deteriorating the overall performance of the network.
To this end, we propose a novel multi-path hybrid
routing protocol, Multipath Mesh (MMESH), that
effectively discovers multiple paths. We also propose
elegant traffic splitting algorithms for balancing traffic
over these multiple paths to synergistically improve the
overall performance. Through extensive simulations, we
observe that our protocol works very well to cope with Figure 1. A sample mesh network in an University
the variations in the network. Our protocol also
improves the performance of flows traversing multiple With the envisaged application of WMN in
hops. providing broadband wireless internet access to
end users, the network should have high capacity
1 Introduction and enough bandwidth to support the
Wireless Mesh Networks (WMNs) are requirements. A wide range of applications with
envisioned to support the wired backbone with a different requirements can be run over these
wireless backbone for providing internet networks, for instance, audio-video conferencing
connectivity to residential areas and offices. The that require minimal jitter and finite delay
primary advantage of these networks is their bounds, file sharing that require large bandwidth
rapid deployment and ease of installation. Unlike etc. amongst others. These applications are
traditional WiFi networks where each access predominantly Internet oriented and thus the
point is connected to the wired network, WMNs traffic is either from the end users towards the
use the paradigm of multihop communication. In Internet gateway or in the reverse direction [1].
WMNs, Mesh Routers (hybrid version of Access This is a sharp contrast to MANETs, where
Points) communicate with the external network traffic can be assumed to be uniformly
(for e.g. Internet) by cooperatively forwarding distributed between any pair of nodes. Intuitively,
each others’ traffic towards the gateway nodes any routing algorithm designed for a WMN will
(a.k.a Mesh Points) which are directly connected aim to invariably choose the best links while
to the wired backbone. In this paper, we use the finding routes from any router to IGW.
terms APs and MRs interchangeably. Similar to Consequently, certain nodes or links can be
the wired network where intermediate routers heavily loaded while some nodes/links are
form backhaul and route traffic from one network seldom used. However this may lead to an
to another, in a WMN, Mesh points and Mesh undesirable situation in which the best paths may
routers forward each other’s traffic in order to degrade due to the load, consequently resulting in
establish connectivity. Note here that Mesh suboptimal performance. Even a routing protocol
routers (MRs) and Mesh points (MPs) are similar that periodically monitors links/paths cannot help

1-4244-0507-6/06/$20.00 ©2006 IEEE 741


in such a case, because it may result in too many However they do not consider load balancing i.e.
oscillations and lead to an unstable operation of the traffic concentration and congestion on
the network. Moreover, in a WMN, the traffic certain paths. Two nodes may choose the same
that is heterogeneous in nature and comprising of next hop node after hearing an advertisement,
significant multimedia content, supporting the thus leading to potential bottleneck nodes (hot
desired Quality of Service (QoS) becomes an next-hop). Another problem is if a node has good
important requirement. Thus, recalling the chief link to the next hop but it is heavily loaded then
functionality of a WMN as the means of there is no use in selecting a path to IGW through
extending internet connectivity, any proposed such a node. Thus their protocol is may not be
solution must efficiently balance the traffic in the scalable to the size of a community WMN (which
network and yet meet the application demands. envisions providing wireless internet connectivity
In a WMN, it is possible to reach an IGW or to users). Yang et al. [5] showed that WCETT
any mesh router (MR) through multiple paths. can in fact create routing loops in certain
Thus if multiple paths can be utilized to balance situations. They further propose another routing
the traffic, then the aforementioned situation protocol (LIBRA) considering the intra and inter
would seldom occur. In this paper, we propose to flow interference. Even this protocol does not
exploit multiple paths to synergistically improve focus on traffic congestion buildup along good
the overall performance of WMNs. To this end, paths. Raniwala et al. [4], explore distributed
we propose a novel multi-path route discovery channel assignment and routing in a multi-
and maintenance for WMN. We further propose a channel WMN by creating a spanning tree rooted
traffic splitting algorithm that can be used over at the gateway, but they do not focus on network
these multiple routes to balance the traffic load load balancing. Ramachandran et al. [6] propose
uniformly. Rest of the paper is organized as another spanning tree based protocol AODV-ST
follows: Section 2 reviews some related work on that modifies the AODV protocol to include
routing protocols for mesh networks. In section Expected Transmission Time (ETT) as the
3, we describe our proposed multipath routing routing metric. Each AP routes traffic through a
protocol. We then present some traffic splitting primary gateway that has least end-to-end delay.
algorithms for efficiently balancing the load over In order to estimate the least loaded gateway,
different paths in section 4. In section 5, we study periodic RTT probing is performed which is
the performance of our protocol. Section 6 expensive. Moreover they do not consider
presents some conclusions and future directions routing in multi-channel architecture, thus inter-
of research. flow and intra-flow interference still remain a big
challenge. Alicherry et al. [9] propose heuristics
2 Related Work for joint channel assignment, routing and link
Using traditional routing protocols designed for scheduling problem for multi-radio wireless
MANETs may actually create many hotspots, mesh networks. They focus on achieving optimal
and lead to subdued performance. For example channel assignment with an interference-free link
popular routing protocols for Ad Hoc Networks schedule and routing by satisfying certain
such as AODV, DSR decide the routes solely fairness, link congestion, link scheduling
based on the number of hops. Recent work [2] [8] constraints among others.
has revealed that shorter hop routes prevent the
network from realizing higher performance when 3 Multi-path routing in wireless mesh
the wireless interface has multi-rate capability. networks (MMESH)
They present a source routing protocol (LQSR) As explained earlier, routing protocols designed
[3] for multi-radio multi-channel WMNs using a for MANETs primarily focus on finding a single
novel routing metric, Weighted Cumulative best possible route to any destination. However
Expected Transmission Time (WCETT). The in WMNs, traffic is primarily routed either
proposed metric enables the nodes to choose the towards the IGWs or from the IGWs to the APs.
best possible routes that have good mix of Thus if multiple APs choose the best path
channel variant hops and high bandwidth links. towards a gateway, the traffic in this path

742
increases. As more traffic is routed through a best through this notification message, a child MR
route, it increases the load on the intermediate notifies to its parent MR which paths to use for
MRs and thereby deteriorating the overall forwarding its traffic. Parent MRs receiving this
performance of the path. As can be noticed this notification message then registers the child MR
performance degradation is due to the traffic in its routing table and updates the route(s) that
build up on certain hot paths. But, if we utilize should be used to forward the traffic from the
multiple paths to balance the traffic, such a child MR. The parent MR also establishes a
situation would seldom occur. We propose to reverse route to the child MR.
exploit these multiple paths to synergistically After processing the
improve the overall performance in a WMN. PARENT_NOTIFICATION message, a parent
MR will then unicasts another notification
3.1 Multi-path Route Discovery
message (called CHILD_NOTIFICATION) to all
Each node in the network selects a set of the corresponding MRs (that occur in the selected
routes from itself to the destination and informs routes). This CHILD_NOTIFICATION is
all the corresponding intermediate nodes. The propagated all the way until it reaches the IGWs.
idea of restricting the set of routes for forwarding This CHILD_NOTIFICATION message informs
traffic from a particular source is derived from all the intermediate MRs along the route
the traditional source routing technique which including the IGW about a child MR and the path
offers a unique advantage. With source routing, that can be followed to reach this child MR. On
we can incorporate the decision of choosing the receiving the CHILD_NOTIFICATION message,
route with good channel diversity which is each parent MR registers this child MR and
otherwise not possible in distance vector routing. follow similar steps (as described earlier) in
We divide the route discovery and route registering the multiple route(s) to reach a child
maintenance procedures into different phases. MR in their respective routing tables. Thus each
The following subsections will describe the intermediate MR (including the IGW) that is in
details of the protocol. the path from a child MR to IGW now has one or
3.1.1 Initial Network setup phase more route(s) to the corresponding child MR.
This phase allows all the nodes to discover We now illustrate the operation of our
multiple routes to the IGWs. The IGWs and MRs protocol using a simple example. Consider the
that have at least one path to the IGW can sample mesh network shown in figure 2. In this
broadcast the advertisements. Initially, when the figure there are two gateway nodes (GW0 and
network is just started, only the gateway nodes GW1) which are connected to the wired network.
broadcast advertisements of internet connectivity. These gateways announce their presence by
Neighboring MRs setup paths to all the possible periodic beacons which include some additional
IGWs in the order of their performance. These information as discussed earlier. MRs 1, 3 and 4
nodes further broadcast the connection receive these advertisements. MR 3 receives two
advertisements. In these advertisements they advertisements (from GW 0 and GW 1). Let us
announce all their routes to the IGWs with their suppose that the link between MR 3 and GW 1 is
order of preference. These advertisements can better than the link between MR 3 and GW 0. It
include certain metrics such as ETT or delay to thus prioritizes its routes to the GWs according to
the Gateways (additional information such as its the link capacity and path performance (WCETT
local load, average queue lengths etc. can also be can be used as a metric).
included). A MR after hearing these Now consider the case of MR 6. Let us suppose it
advertisements then decides the different routes receives the first route advertisements to GW
that are acceptable to it and adds this to its from MR 5 (5-4-GW0, and 5-1-GW0). Since it
routing table. It then unicasts a doesn’t have any routes already, it selects both of
PARENT_NOTIFICATION message to the them and sends a PARENT_NOTIFY message to
selected parent MRs (from whom it has selected MR 5 including its preference of the routes to
the route(s)). This message contains all the routes MR 5. MR 5 then creates an entry in its routing
it has chosen from the advertised routes. Thus table and notes the preferred next hops as
specified by the child MR (MR 6). It then sends

743
a CHILD_NOTIFY message to notify its parents 3.1.3 Fast rediscovery of paths
(in this case MR 4 and MR 1) about the child Although topology changes in a WMN are very
(MR 6) along with the selected routes. Similar to uncommon, nevertheless there is still some
MR 5 they also register the preferred next hops possibility. Some possibilities are when a new
for traffic from MR 6. This notification goes until MR is added to the network or a MR reboots.
it reaches the GW0. Moreover due to varied reasons 1 , the links are
highly unstable and temporary glitches can occur.
In such cases, some routes may become stale and
thus it is highly important for any routing
algorithm to quickly adapt to these changes by
facilitating faster recovery.
We consider the two cases when an existing
MR reboots or goes down in an active network.
In such a case, the adjacent nodes will detect its
absence and update the routing table. Specifically
when an MR detects the failure of a neighbor, it
checks its routing table to verify if any route has
this neighbor in the next hop field. If such routes
Figure 2: Mesh network scenario are found, they are temporarily suspended. When
a route is suspended, the traffic from the affected
3.1.2 Route maintenance source nodes is routed through other possible
A WMN is an ad hoc network and hence can next hop nodes. If however there was only one
undergo periodic up gradation. As new MRs can next hop chosen by the source node, then the MR
be added or some existing MRs may be switched is authorized to route the traffic through alternate
off etc, new routes should be discovered and paths until the threshold time. If the neighbor
some obsolete routes should be detected. The node recovers before this threshold time, the
MRs should be promptly informed about these route is made active again. Otherwise the source
changes in the network and avoid sending traffic node will be notified by an RERR.
over stale routes.
In the route maintenance phase of our 4 Traffic balancing at Mesh Routers
protocol, we facilitate the monitoring and In the previous sections, we have seen how each
exchange of the changing information in the MR (and Gateways) elegantly discovers and
network. After initial route discovery phase, each maintains multiple routes to and from the
MR continuously monitors the performance of all Gateways. Once we have these multiple routes
the active paths. On finding new routes or stale setup, our task is to divide the traffic among these
routes, a MR appropriately validates its routing routes to balance the load, without degrading the
tables and announces this information to its performance. In this section, we will describe
neighboring nodes. Neighboring nodes promptly how each MR uses these available route(s) to
update their routing table using this information achieve load balancing.
and further propagate the changes.
In order to facilitate faster failure recovery, 4.1 Round Robin scheduling
we provide certain authority to intermediate This is by far the simplest traffic splitting
nodes. When an MR detects the failure of a mechanism that can be used for effective traffic
particular next hop, it immediately checks its load balancing in a network. A node sends every
routing table to identify all the routes that pass packet to a different next hop thus uniformly
through the failed MR. Upon identifying all such distributing traffic over in the network. However
routes; it temporarily suspends the scheduling this scheme suffers from setback due to possible
through this node. Thus, even though the source out of order delivery of packets belonging to
nodes that have selected the routes that are far
away, they are not affected by any intermediate 1
Possible reasons are due to small scale fading and
node failures. multipath signal transmissions

744
same flow. The transport layer protocols need to attributed due to the path selected by MR 8 for
maintain a large buffer to cope up with out of reaching the IGW. As it uses only a single path,
order delivery and can lead to unnecessary loss traffic from nearby nodes can choke the longer
with protocols such as TCP. hop length flows [7]. Self interference and unfair
channel access are the other reasons that
4.2 Congestion aware routing
contribute to this starvation of this client.
One of the important features of any multi-path On the other hand, when we use our MMESH,
routing algorithm should be the ability to the traffic is distributed over multiple routes.
effectively mitigate congestion and avoid routing Figure 3(b) shows the throughput performance
traffic through congested routes. High average when MMESH is used. We observe a substantial
queue lengths indicate high congestion at a node improvement in the throughputs of flows from
which can lead to possible packet droppings in MR 8. Since the packets are distributed over
turn decreasing the performance of the network. multiple routes, the packets that are transmitted
Thus if we find that a particular next hop node through the non-congested nodes are thus not
has high average queue length over a period of dropped due to overflowing buffers. We observe
time, we temporarily skip that node and send the that a majority of these packets can successfully
traffic through other neighboring nodes. reach the IGW. However as we still route some
of the packets through the congested nodes, these
5 Performance Evaluation are lost due to the buffer over flows.
In this section, we evaluate the performance Figure 3(c) shows the instantaneous
of our proposed routing protocol using extensive throughputs of the three flows when we use our
simulations using ns-2. We have implemented congestion aware multipath routing mechanism.
our MMESH protocol in ns-2 simulator (version We can notice the increased throughput of the
2.29) [9]. We also enhanced ns-2 by modifying flow from MR 8. Recall from section 3.2, each
its default 802.11 MAC layer implementation to node reports some of the node specific metrics
support the multi-rate transmissions specified by such as average queue length to its neighbors
IEEE 802.11b standard. We considered a 3x3 through the periodic HELLO advertisements.
grid of MRs serving up to 50 end users/clients This information is used to determine whether a
(CN). Each MR is equipped with two radios: one packet should be sent or skipped through a
for communicating in the backbone network i.e. particular next hop node. When a MR notices
with other peer MRs while the second radio is high average queue lengths in a next hop node
used to explicitly communicate with CNs. The for a period of time, it avoids sending any
bottom left MR i.e. MR0, serves as the Internet packets through this node until the congestion
gateway for all other MRs. reduces. This optimization not only helps in
5.1 Throughput performance with UDP traffic
improving the performance of the flows by
reducing the packet drop probability, but also
We start aggregate of flows from CNs under greatly reduces the wastage of valuable
MR 8, MR 6, and MR 4 when the simulation bandwidth.
time is 5, 20, and 30 seconds respectively. Figure Next we also measure the throughputs of these
3 (a-c) shows the instantaneous throughput of the clients by varying the traffic generation rates of
flows from the three clients. As we can observe, their flows. Figure 4(a-c) shows the throughput
when we use AODV as the routing protocol, the with increasing traffic generation rate. We can
clients that are closer to the IGW enjoy very high observe a tremendous improvement in the
throughput. Initially when only CNs under MR 8 throughput of clients under MR 8 without
are transmitting in the network, they achieve affecting the performance of other clients. From
decent throughput of around 300 Kbps. But when figure 4 (c), we can notice that Congestion aware
the flows from CNs under MR 4 start, flows from MMESH helps in further improving the
MR 8 suffer from severe packet loss. performance by as high as 100%.
Different factors contribute for this poor
performance. The primary reason for this can be

745
350 MR-8 400 MR-8 400 MR-8
MR-6 MR-6 MR-6
300 350 350
MR-4 MR-4 MR-4
300 300
Throughput (Kbps

Throughput (Kbps
Throughput (Kbps
250
250 250
200
200 200
150
150 150
100
100 100
50 50 50
0 0 0
0 25 50 75 100 125 150 0 25 50 75 100 125 150 0 25 50 75 100 125 150
Time (s) Time (s) Time (s)

(a)AODV (b) MMESH (c) Congestion aware MMESH


Figure 3: Instantaneous throughput of the flows observed with different routing protocols

350 250 250

300
200 MR-8 200
Throughput (Kbps

Throughput (Kbps
Throughput (Kbps

250 MR-8 MR-6


150 MR-4 150
200 MR-6
MR-4
150 100 100
100 MR-8
50 50 MR-6
50 MR-4
0 0 0
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
Load (Kbps) Load (Kbps) Load (Kbps)

(a)AODV (b) MMESH (c) Congestion aware MMESH


Figure 4: Aggregate Throughput of the flows from clients under different mesh routers

[2] R. Draves, J. Padhye and B. Zill, “Comparison of Routing


Metrics for Static Multi-Hop Wireless Networks,” Proc of
6 Conclusion SIGCOMM 2004
In this work, we propose a proactive multi-path [3] ---, “Routing in Multi-Radio, Multi-Hop Wireless Mesh
routing protocol which exploits the multiple routes Networks,” Proc of MOBICOM 2004
between the MRs and the gateways to balance traffic [4] A. Raniwala, and T. Chiueh, “Architecture and Algorithms
for an IEEE 802.11-based Multi-channel Wireless Mesh
load. We also propose different techniques to balance Network,” Proc of IEEE Infocom '05.
the traffic over these multiple paths. Simulation [5] Y. Yang, J. Wang and R. Kravets, “Designing Routing
results reveal that single path routing leads to severe Metrics for Mesh Networks,” IEEE Workshop on Wireless
packet loss and consequently very low end –to-end Mesh Networks, WiMesh, 2005
[6] K. Ramachandran et al., “On the Design and
throughput for longer hop length flows. Our protocol Implementation of Infrastructure Mesh Networks,” IEEE
helps in dramatically increasing the performance of Workshop on Wireless Mesh Networks (WiMesh), 2005.
the longer hop length flows. Even with naïve round [7] N. S. Nandiraju, D. Nandiraju, D. Cavalcanti, D. P.
robin scheduling, we are able to substantially Agrawal, “A Novel Queue Management Mechanism for
increase the throughput. Congestion aware routing Improving Performance of Multihop Flows in IEEE 802.11s
based Mesh Networks,” in the Proc of IPCCC, April 10-12,
improves the throughput by almost 100% compared 2006.
to the naïve round robin scheduling. As a part of [8] D. D. Couto, D. Aguayo, J. Bicket, and R. Morris, “A High-
future work we plan to evaluate our protocol with Throughput Path Metric for Multi-Hop Wireless Routing,”
multiple radios in the backbone network. Proc. of ACM MOBICOM, 2003.
[9] M. Alicherry, R.Bhatia, L.Li, “Joint Channel Assigmnet and
Routing for Throughput Optimization in Multi-radio
References Wireless Mesh Netwtorks,” Mobicom 2005.
[1] N. Nandiraju, D Nandiraju, L Santhanam, B He, J.F. Wang, UCB/LBNL/VINT Network Simulator (NS-2), Available at
D. P. Agrawal, “Wireless Mesh Networks: Current http://www.isi.edu/nsnam/ns/index.htm
Challenges and Future Directions of Web-in-the-sky, ” To [10] UCB/LBNL/VINT Network Simulator (NS-2), Available at
Appear in IEEE Wireless Communications Magazine, 2006 http://www.isi.edu/nsnam/ns/index.htm

746

Das könnte Ihnen auch gefallen