Computer Networks 51 (2007) 3142–3156 www.elsevier.com/locate/comnet
Routing and quality of service support for mobile ad hoc networks Anelise Munaretto
a,c,* ,
Mauro Fonseca
b,c
a
CPGEI/UTFPR, Programa de Po´s-graduac¸a˜o em Engenharia, Ele´trica e Informa´tica Industrial, Av. 7 de setembro, 3165, 80230-901 Curitiba (PR), Brasil b PPGIA/PUC-PR, Bloco 2 – Parque Tecnolo´gico – 2 andar, Rua Imaculada Conceic¸a˜o, 1155, Prado Velho, CEP-80215-901 Curitiba (PR), Brazil c LIP6/Universite´ de Paris VI, 104 avenue du Pre´sident Kennedy, 75016 Paris, France
Received 30 November 2005; received in revised form 17 December 2006; accepted 28 December 2006 Available online 1 February 2007 Responsible Editor: V.R. Syrotiuk
Abstract OLSR is an optimization over classical link state protocols tailored for mobile ad hoc networks. In this paper, we propose the QOLSR protocol which includes quality parameters to the standard OLSR. Three variants of QOLSR are introduced, taking into account the delay measurement together with the hop count metric. Then, we analyze new heuristics for the multipoint relay selection, and evaluate our proposed approaches comparing them with the standard OLSR protocol. Simulation results show that an increased load-balancing and a reduced dropped packets rate are achieved due to the inclusion of the delay information. 2007 Elsevier B.V. All rights reserved. Keywords: Ad hoc networks; Routing protocols; QoS support; Delay metric
1. Introduction The highly dynamic nature of a mobile ad hoc network results in frequent and unpredictable changes in the network topology, increasing the complexity of routing among nodes. Such chal*
Corresponding author. Address: CPGEI/UTFPR, Programa de Po´s-graduac¸a˜o em Engenharia, Ele´trica e Informa´tica Industrial, Av. 7 de setembro, 3165, 80230-901 Curitiba (PR), Brasil. Tel.: +55 41 3310 46 86. E-mail addresses:
[email protected] (A. Munaretto),
[email protected] (M. Fonseca).
lenges make routing probably the most active research topic within the MANET area. Besides the challenges associated with mobility, more difficulties are introduced by the specific characteristics of the wireless channel. Broadcast is the basic mode of operation over a wireless channel where, in general, each transmitted message can be received by all neighbors located within one-hop from the sender. In terms of traffic classification, in unicast the MAC layer is supposed to filter the messages and deliver them to higher layers those whose address matches with the node. When, broadcast is used all neighbors that receive
1389-1286/$ - see front matter 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.comnet.2006.12.010
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
the message will forward them to higher layers regardless of its address, while the routing will forward the messages to other nodes. Such broadcast traffic is often used by routing protocols for path discovery. In this case, the simplest implementation of the broadcast operation is by naive flooding, which may cause the broadcast storm problem due to redundant re-broadcast [1]. Numerous routing protocols and algorithms have been proposed. Their performance under various network environments and traffic conditions have been extensively studied and compared [2]. MANET routing protocols are typically subdivided into two main categories: reactive on-demand routing protocols and proactive routing protocols. Reactive on demand routing protocols establish the route to a destination only when there is a demand for it. In proactive routing protocols routes are calculated before needed. Such protocols can be derived from either legacy Internet distance-vector or linkstate protocols. For mobile ad hoc networks, the proactive routing protocols are table-driven, where each node tries to maintain routing information about every other node in the network at all times. An example of a proactive protocol is OLSR [3], which is an optimization over the classical link state protocol [4] (e.g. OSPF [5]). OLSR is now officially defined by the RFC 3626 of the IETF [6] in the Mobile Ad hoc NETworks (MANET) working group. It performs hop-by-hop routing, i.e., each node uses its most recent information to route a packet. Therefore, each node selects a set of its neighbor nodes as MultiPoint Relays (MPRs) [7]. In the OLSR protocol, only the nodes selected as MPRs are responsible for forwarding control traffic, intended for diffusion to the entire network. MPRs provide an efficient mechanism for flooding control traffic that reduces the number of required transmissions. The MPRs are also responsible for declaring the link state information over the network. However, no QoS information is taken into account, leading to a non optimal path selection in terms of QoS requirements. In this paper we propose a QoS-enhancement for the OLSR protocol, which we define as the QOLSR protocol. The QOLSR protocol extends the standardized OLSR protocol [3], introducing QoS metrics to wireless and mobile ad hoc networks. While the hop distance may be a valid metric for wired and stationary networks, it does not consider the specifics of wireless links nor node movement. We
3143
introduce three QOLSR variants of OLSR with different tradeoffs: QOLSR1, QOLSR2 and QOLSR3. The three variants – QOLSR1, QOLSR2 and QOLSR3 are heuristics designed for the selection of MPRs based on QoS parameters. The heuristic used in QOLSR1 selects as MPR the neighbor node that can reach the largest number of nodes (such a node will have what we define as the maximum reachability or degree). If there are two or more neighbor nodes with the same reachability, then QOLSR1 prioritizes the neighbor with the smallest delay. The next heuristic is the one used in QOLSR2, which prioritizes the node with the smallest delay when selecting the MPR node. If there are two or more neighbors with the smallest delay, then the node to be chosen as MPR will be the one with the largest reachability. The last proposed heuristic, used in QOLSR3, selects as MPR the neighbor node with the smallest delay among the neighbors that are, at most, within two hops from the initial node. We demonstrate that QOLSR3 finds the optimal shortest path, in terms of delay, using only partial knowledge of the network topology. The remainder of this paper is organized as follows. A detailed specification of the QOLSR protocol is presented in Section 2, while in Section 3 we present its performance evaluation. In Section 4 we conclude the paper. 2. QOLSR 2.1. OLSR OLSR [3] is a proactive routing protocol that shares the stability of link state algorithms [4] and the advantage of having the routes immediately available when needed. In pure link state protocols, all links within neighbor nodes are declared and control messages are flooded over the entire network. The OLSR protocol is an optimization of pure link state protocols (e.g. OSPF [5]) for mobile ad hoc networks. First, it reduces the size of the control packets: instead of all links, it declares only a subset of links within those neighbors that are in the MPR set [8]. Second, it minimizes the flooding of control traffic by using only those nodes within the MPR set to broadcast its messages. Therefore, only MPRs of a node rebroadcast packets. This technique significantly reduces the number of retransmissions in a flooding or broadcast procedure [7]. A detailed description of the OLSR protocol can be found in the RFC3626 [3].
3144
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
2.2. Problem statement In the literature, as in [9], compulsory or guaranteed services are sometimes also referred to as predicted or deterministic services, respectively. Deterministic service offers hard QoS guarantees to all packets belonging to a data stream. Predicted service is based on statistical bounds and provides weaker QoS guarantees than deterministic service, but stronger ones than best-effort service. Providing a QoS service as opposed to a best-effort service is a very complex problem in MANETs [10]. A network’s ability to provide QoS depends on the intrinsic characteristics of all network components, from transmission links to the MAC and network layers [11]. Such particular MANET characteristics generally lead to the conclusion that this type of network provides a weak support to QoS. Wireless links have time-varying capacity and larger loss rates than wired links. Topologies are highly dynamic with frequent link outage. Random access-based MAC protocols commonly used in this environment have no QoS support, e.g. the IEEE 802.11 family. Moreover, MANET link layers typically run in unlicensed spectrum, making it more difficult to provide strong QoS guarantees. This scenario indicates that hard QoS is extremely difficult to guarantee in a MANET. If the nodes have high mobility then even statistical QoS guarantees may be hard to achieve due to the lack of sufficient accurate knowledge of the network states. There is a growing consensus [12,13] that adaptive quality of service models present the only viable approach to addressing the technical challenges associated with wireless networks. A more realistic approach for QoS provisioning in an ad hoc network is based on an adaptive QoS model where applications must adapt to the time-varying nature of the network. In [14], the definition of the QoS model for a MANET includes a set of parameters that can adapt an application to the current quality of the network. The proposed QOLSR protocol introduces quality parameters to the standard OLSR protocol. Since the quality parameters are available before the routing table calculation, the QoS constraints can be verified before the route selection. The applications can be adapted really knowing the quality that is expected to be found over the network. Therefore, an adaptive cross-layer QoS model, including the application layer, must be proposed to adapt the applications. However, this model is out of the scope of this work. An adaptive
QoS routing protocol providing more appropriate metrics is already a considerable improvement in terms of achievable performance for mobile and wireless environment. In order to design a QoS routing protocol for MANETs based on a link state approach one should: (i) mitigate control traffic, which is the main drawback of proactive protocols; (ii) avoid oscillations and instabilities, which are typically generated by frequent changes on selected routes; (iii) improve overall performance when compared with the standard metric based on hop count. 2.3. Our proposal No additional control messages are included for QOLSR. We benefit from Hello and TC messages to evaluate and maintain QoS information through the network. Nevertheless, to carry this QoS information, the format of Hello and TC messages must be changed, which increases the size of these messages. In spite of that, the final control traffic load is smaller than if new control messages were included. In Section 3 we evaluate the proposed protocol in terms of control traffic. The QOLSR protocol aims to introduce more appropriate metrics for mobile ad hoc networks. Many QoS parameters, e.g. bandwidth, one-way delay, end-to-end delay, energy, cost, jitter, packet loss probability, stability, etc., can be relevant in analyzing a wireless and mobile environment. However, the impact of each QoS parameter depends on the application requirements. We present how we measure these metrics and how these measurements are used as QoS metrics by QOLSR. These measured values are included in the QOLSR protocol functionalities. QOLSR does not require any changes in the format of IP packets. Thus, any existing IP stack can be used and the protocol only interacts with routing table management. As in standard OLSR, link state information is generated only by nodes selected as MPRs. This information is then used for route calculation in all nodes. 2.4. Metric measurements This section presents QoS metrics and how we have designed the measurement of these metrics. 2.4.1. Delay metric The first QoS parameter to be analyzed is the oneway delay metric. Due to the fact that Hello messages are ‘‘broadcast’’ in a controlled manner through the
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
network, there is no acknowledgement of broadcast messages. Consequently, we consider one-way delay rather than end-to-end delay. The measurement of the one-way delay avoids the increase of traffic load typically imposed by the addition of extra messages to the QOLSR protocol in order to estimate the end-to-end delay. However, for successful measurement of the one-way delay, a single global time axis is required. Clocks in ad hoc networks, wireless networks and sensor networks, when synchronized via GPS [15], NTP [16], or any of many efficient synchronization protocols for wired as well as wireless media [17–19], allow the assumption about an approximate, but sufficient, single global time axis. Thereby, based on the above considerations, we assume synchronized clocks. Finally, the assumption of a synchronized network simplifies the one-way delay calculation in the ubiquitous environment. During the neighbor discovery performed by QOLSR, each node generating a Hello message includes its creation time. When a neighbor node receives this message, it calculates the difference between such a time and the current time, which represents the measured one-way delay. This measured one-way delay includes the queuing time, the transmission time, the collision avoidance time, and the control overhead time. Moreover, such a measurement can drastically vary over time. However, it represents a relative measurement since is done using broadcast messages rather than unicast messages as used by data packets. Therefore, we opt for an estimate value avoiding the increase of the traffic control. Furthermore, we use the weighted average method to accommodate varying delays by using an adaptive algorithm. Our method is based on the RTT (Round-Trip-Time) estimation in ‘‘congestion avoidance and control’’ [20] for TCP [21], which is useful for calculating weighted averages across multiple intervals.
3145
short time. Otherwise, if a 1 then the formula makes weighted average respond quickly to delay changes. In order to make a tradeoff between delay variations and path oscillations, we define a threshold value, e.g. 10%. Then, a new average_delay is only computed if and only if the last measurement, i.e. measured_delay, is more than such a threshold value. Otherwise, we conserve the oldest average delay without changing the delay measurement. More discussion about the path oscillation problem is presented in Section 2.7. 2.4.2. Bandwidth metric The bandwidth measurement is dependent on the link layer technology. The IEEE 802.11 technology is widely used in testbeds and simulations for wireless ad hoc network research. Nevertheless, ad hoc networks present even greater challenges than infrastructure wireless networks at the MAC layer [10]. For the sake of simplicity, and for calculating the available bandwidth, we consider the medium access control scheme described in IEEE 802.11b. The method proposed in this section calculates the available bandwidth between a given node and its neighbors based on the work of Gerla and coworker [22], which by its turn considers the acknowledgement time of the data packets for measuring the bandwidth metric. Our proposed method includes both data packets and signalling traffic (e.g. Hello messages and TC messages in the OLSR protocol), since they also use the available bandwidth and must be taken into account. Suppose node i and its neighbor node j, then we define the available bandwidth between them as follows: Bwði;jÞ ¼ ð1 uÞ Throughput packetði;jÞ ;
average delay ¼ a old average delay þ ð1 aÞ measured delay
where u is the link utilization. The throughput seen by one packet of S bits can be calculated as:
where
Throughput packet
• average_delay: represents the new weighted average, • old_average_delay: stores a weighted average, which is slowly changed based on measured_ delay, • measured_delay: computed from the elapsed time between sending a Hello message and receiving it, • 0 6 a 6 1: if a 0 then the formula makes weighted average immune to changes lasting a
¼
S PR tq þ ðtS þ tCA þ toverhead Þ R þ r¼1 BT
where tq is the MAC queuing time, tS the time to transmit S bits, tCA the collision avoidance time, toverhead the control overhead time (e.g. RTS, CTS, etc), R the number of necessary transmissions, BT the backoff time for retransmission r. However, as shown in [22], this formula reveals some undesirable characteristics such as packet size
3146
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
dependence and high variance due to random per packet effects. To increase the statistical robustness of the measurements, a packet window of 16 or 32 samples (packets) is shown to be adequate. To illustrate, authors in [22] show where the high variance per packets measurements are aggregated on a window of 32 packets. In this case, the Link_throughput is the aggregated measurement of the Throughput_packet. The idle_time is the duration time during which a node is idle and the window_duration represents the window size. The idle_time and window_duration are calculated to produce the link utilization factor and the permissible throughput measurement as: Permissible throughput idle time ¼ Link throughput window duration The ad hoc network size is used to classify MANETs [10]. The scale of an ad hoc network can be classified as small-scale (i.e., 2–20 nodes), moderate-scale (i.e., 20–100 nodes), large-scale (i.e., more than 100 nodes), and very large-scale (i.e., more than 1000 nodes). Many works [23] have presented possible manners to calculate the bound on throughput for ad hoc networks. However, the achievable throughput can change due to network conditions [24]. Moreover, the available bandwidth undergoes fast time-scale variations due to channel fading and error from physical obstacles [25]. These effects are not present in wired networks. Then, to make estimation of available bandwidth and accurate throughput calculation in wireless networks are challenging tasks. Furthermore, the wireless channel is also a shared-access medium, and the available bandwidth also varies with the number of hosts contending for the channel and their bit rates [26]. As a result of these challenges and due to the instability of the bandwidth value in IEEE 802.11 networks, we perform our simulations presented in this paper, considering only the delay metric rather than the bandwidth metric. We extend upon earlier work in [27] by using a more realistic measurement of the quality of the network. 2.5. Selection of multipoint relays This section discusses and presents QoS-based MPRs selection heuristics. We start defining the terminology used and presenting the standard MPR selection defined in RFC3626 [3]. Afterwards, we
proposed new heuristics considering QoS parameters when selecting MPRs. 2.5.1. Terminology According to the RFC3626 and based on [28,29], if we consider delay metrics, the following terminology will be used in describing QOLSR algorithms and heuristics: • Neighbor of an interface: a node is a ‘‘neighbor of an interface’’ if the interface (on the local node) has a link to any one interface of the neighbor node. • 2-hop neighbors reachable from an interface: the list of two-hop neighbors of the node that can be reached from neighbors of this interface. • MPR set of an interface: a (sub)set of neighbors of an interface selected such that through these selected nodes, all strict two-hop neighbors from that interface are reachable. • N(x): N(x) is the subset of neighbors of the node x, which are neighbors of its interface. • N2(x): the set of two-hop neighbors reachable from the node x. • D(x,y): the degree of a one-hop neighbor node y where y is a member of N(x), and is defined as the number of symmetric neighbors of node y, excluding all members of N(x). • Shortest path: a path with minimum delay, calculated by the source node based on its known partial network topology. • Optimal shortest path: the shortest path between two nodes in the whole network topology. Any node in the network can be selected as an intermediate node in the optimal shortest path. 2.5.2. Standard MPR selection Finding a MPR set with minimal size falls in the category of dominating set problems, which are known to be NP-complete. Demonstrations and proofs were detailed in [7]. The information needed to calculate MPRs is the set of one-hop neighbors and two-hop neighbors. The proposed heuristic in [3] to calculate multipoint relay set of node x is as follows: • Step 1: Start with an empty multipoint relay set MPR(x). • Step 2: Calculate D(x,y), "nodes y 2 N(x). • Step 3: First, select those one-hop neighbor nodes in N(x) as multipoint relays which provide the ‘‘only path’’ to reach some nodes in N2(x), and
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
add these one-hop neighbor nodes to the multipoint relay set MPR(x). • Step 4: While there still exist some nodes in N2(x) that are not covered by the multipoint relay set MPR(x): – Step 4a: For each node in N(x) which is not in MPR(x), calculate the number of nodes that are reachable through it among nodes in N2(x) and which are not yet covered by MPR(x). – Step 4b: Select the node from N(x) as a MPR that reaches the maximum number of uncovered nodes in N2(x). • Step 5: To optimize, remove each node in MPR(x), one at a time, and check if MPR(x) still covers all nodes in N2(x).
3147
x v
y
a
b
s j
z
c
m i t
d
h
w
e g f
r
q k
The third step permits to select some one-hop neighbor nodes as MPRs which must be in the MPR(x) set, otherwise MPR(x) will not cover all two-hop neighbors. So these nodes will be selected as MPRs in the process, sooner or later. The heuristic used in the standard OLSR protocol computes a MPR set of cardinality at most log n times the optimal multipoint relay number, where n is the number of nodes in the network [30]. The standard OLSR heuristic limits the number of MPRs in the network, ensuring the overhead to be as low as possible. However, in QoS routing, by such a MPR selection mechanism, the good quality links may be hidden to other nodes in the network. There is no guarantee that OLSR finds the optimal shortest path with respect to the delay metric. By example. From Fig. 1 we construct the Table 1 considering the standard OLSR heuristic: Based on the proposed heuristic, node m will select its MPRs as follows: • The Step 3 of the heuristic does not apply because all two-hop nodes are reachable through more than one one-hop neighbor. • Going to Step 4 of the heuristic, we have a tie, because node a and node j have the same reachability and the same degree. Then, we suppose that the node a is randomly selected as an MPR. Therefore, MPR(m) = a. Then, the nodes from N2(m) which are now covered by a node in the MPR set are removed. The Step 4 is repeated while there exist nodes in N2(m) which are not covered by at least one node in the MPR set.
Node selecting MPR
One-hop neighbor
2-hop neighbor
Fig. 1. Example for the multipoint relay selection.
Table 1 Example for multipoint relay selection Node
One-hop neighbor
Two-hop neighbor
MPR
m
a, b, c, d, e, f, g, h, i, j
s, v, x, y, z, w, q, k, r, t
a, e, h, c
• Finally: MPR(m) = a, e, h, c. Now add constraints to the graph in Fig. 1; these are depicted in Fig. 2. When m is building its routing table for destination x using the classical MPR heuristic, it will select the route (m, a, x) whose bandwidth is 15 kbps and the delay is 130 ms. The optimal shortest-widest path between m and x is (m, j, x). It has 280 kbps as bandwidth and 24 ms as delay. The decision on how each node selects its MPRs is essential to determine the optimal QoS route in the network. In the MPR selection, links with best QoS resources should not be omitted. We present in this section three heuristics for the MPR selection: QOLSR1, QOLSR2 and QOLSR3 considering the minimum delay path. 2.5.3. QOLSR1 This heuristic is similar to the standard MPR selection except when we have a tie, because node a and node j have the same reachability and the same degree. Then, rather than randomly selecting an MPR, our heuristic chooses the one-hop
3148
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156 x
)
2 0m
(30 0k ,
0k
m)
(10 k, 6 0 m)
(4 0
( 10
) k (4 00
, 10m
m) k, 1 0 ( 50 0
)
10 m
3 m)
m)
)
50
f , 10m
k,
e
k, 2
0m )
0m (100k, 1
)
w
,3 0m )
m)
r
(20k
( 300
k, 1
k 00 (4
d
q
k Node selecting MPR
• Step 4b: Select the one-hop neighbor with the minimum delay path as MPR. • Step 4c: In case of a tie in the Step 4b, select the node from N(x) as a MPR that reaches the maximum number of uncovered nodes in N2(x).
( 80
g ) 0m ,1
0k
(1 0
20
) , 60 m
h
,4 0k (3 0 (10 0k, 9 0m) (1 30 k, 20 m )
z
) 0m ,1
(1
(500k, 10 m)
m) , 10 0k
(70 0k
( 50
0k,
10m ) k,
m
(90k, 5m )
i
(200k, 100 m)
c
(1 5 0
5m )
) 1 0m
) 0m
100
0m )
,1
m)
0k
m) 00 ) ,6 k, 2m k ( 300 (1
t
(28
y
)
(40 0k ,
(9 0 0
j
)
0 k,
(5k , 190 m
b , 20
(1 00 k, 1
4 00m
, ( 4k
a ( 30k
s
( 10k,
( 10
k, 9
) 0m
m)
41
0m) (10k, 1 0
(15k , 1 10 m )
0m )
m)
, (1 k
, 25
, 10
( 1k
( 1k
v
ability or the degree. After, if there is more than one MPR with the same minimum delay path, then the MPR with the maximum reachability should be chosen. Such a heuristic modifies the Step 4.b and adds the Step 4c of the standard MPR selection presented in Section 2.5.2 as follows:
One-hop neighbor
2-hop neighbor
Constraints = (bw, delay)
Fig. 2. Example for the multipoint relay selection considering QoS constraints.
neighbor with the minimum delay path. The heuristic used in QOLSR1 prioritizes the minimum shortest path when selecting the MPR. Such a heuristic modifies the standard MPR selection presented in Section 2.5.2 by adding the Step 4c as follows: • Step 4c: In case of a tie in the Step 4b, select the one-hop neighbor with the minimum delay path as MPR. There is no guarantee that QOLSR1 finds the optimal shortest path with respect to the delay metric. By example. From Fig. 2 we construct the Table 2 considering the QOLSR1 heuristic: Applying this heuristic, when m is building its routing table, for destination y, it selects j, e, c, and h MPRs. It will select the route (m, c, y) whose delay is 50 ms. The optimal shortest path between m and y is (m, b, y). It has 20 ms as delay.
There is no guarantee that QOLSR2 finds the optimal shortest path with respect to the delay metric. By example. From Fig. 2 we construct the Table 3 considering the QOLSR2 heuristic: Applying this heuristic, when m is building its routing table, for destination z, it selects i, h, b, j, e and c MPRs. It will select the route (m, c, z) whose delay is 140 ms. The optimal shortest path between m and z is (m, d, z). It has 100 ms as delay. 2.5.5. QOLSR3 QOLSR3 selects as the MPR the neighbor node the one with the smallest delay ‘‘d’’ among the neighbors that are, at most, within two hops from the initial node. Thus, it modifies Step 4b, includes the new Step 4c, and removes Step 5 from the MPR selection algorithm of the standard OLSR protocol, as presented in Section 2.5.2. The proposed algorithm, including the above modifications, is presented next: • Step 4b: Select the node of N(x) as a MPR, which has the minimum delay path considering N(x) and N2(x). • Step 4c: In case of a tie in the Step 4.b, select the node which reaches the maximum number of uncovered nodes in N2(x).
2.5.4. QOLSR2 The heuristic used in QOLSR2 prioritizes the minimum delay path metric rather than the reach-
Claim 1. Consider the case when each edge is given an arbitrary nonnegative delay (length). In this case, the total delay of this path is defined to be the sum of the delays of its edges. In Fig. 3, we define p[s,d] to be a shortest path from s to d; that is:
Table 2 MPR selected in the QOLSR1
Table 3 MPR selected in the QOLSR2
Node
MPR
Node
MPR
N
j, e, c, h
m
i, h, b, j, e, c
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
d
s
d
d
u1
u2
d
d
...
d
d
vj
vi
v1
...
ui-1
3149
d
d
ui+1
uj-1
d
... d
d
ui
uj+1
uk
d
d
uj
Fig. 3. Optimal shortest path from s to d.
p ¼ ðu1 ; . . . ; ui1 ; ui ; uiþ1 ; . . . ; uj1 ; uj ; ujþ1 ; . . . ; uk Þ is an optimal shortest path, for k P 3. For any intermediate node ui (i 5 1) in p that is not selected as MPR by its previous node ui1, we can find a node vi selected as MPR by ui1 such as the path (u1, . . . ,ui1, vi, ui+1, . . . ,uk) has at least the same total delay. Proof. Let p = (u1, . . . , ui1, ui, ui+1, . . . ,uk), k P 3 an optimal shortest path from u1 to uk, as depicted in Fig. 3. Suppose that on the optimal shortest path, the node ui is not selected as MPR by its previous node ui1. We can assume that for each node on the path, its next node in the path is its one-hop neighbor, and the node two hops away from it is its two-hop neighbor. Based on the basic idea of the MPR selection that all two-hop neighbors of a node should be covered by this node’s MPR set, ui1 must have another neighbor vi, which is selected as its MPR, and is connected to ui+1. Let p 0 = (u1, . . . , ui1, vi, ui+1, . . . ,uk), k P 3. According to the criteria of MPR selection specified on QOLSR3, wherein the metric is the delay over two-hops, then ui1 selects vi instead of ui as its MPR because: delay ðui1 vi uiþ1 Þ 6 delay ðui1 ui uiþ1 Þ
ð1Þ
From Eq. (1) we have delay(p 0 ) 6 delay(p). Based on this assumption, if path p is an optimal shortest path, path p 0 is also an optimal shortest path. Claim 2. There is an optimal shortest path in the whole network such that all the intermediate nodes are selected as MPR by their previous nodes. Proof. By a recurrence. Let p = (s, u1, . . . , ui1, ui, ui+1, . . . , uj, . . . , uk, d), j < k an optimal shortest path as depicted in Fig. 3. i. By Claim 1, the first intermediate node u1 is selected as MPR by source s. Then, we can find a node v1 selected as MPR by s such that
the path p 0 = (s, v 1 , . . . , u i1 , u i , u i+1 , . . . , u j , . . . , uk, d) has the same total delay of the optimal path. Then, p 0 is also an optimal shortest path. So, the source’s MPRs are on the optimal shortest path. ii. We assume that all the nodes u1, . . . , ui1, ui, ui+1, . . . , uk are selected as MPRs by their previous node in the path p. We prove that the next hop node of ui on p is ui’s MPR. Suppose that uj+1 is not an MPR of uj. As above, by using the Claim 1, we can find a node vj+1 selected as MPR by uj such that the path p 0 = (s, u1, . . . , ui1, ui, ui+1, . . . , uj, vj+1, . . . , uk, d) has at least the same total delay of the optimal shortest path; then p 0 is also an optimal shortest path. So, in an optimal shortest path, the (j + 1)th intermediate node is the MPR of the (j)th intermediate node. Based on i. and ii., all intermediate nodes of an optimal shortest path are MPRs of previous nodes. By Claim 2, there is an optimal shortest path such that all intermediate nodes are the MPRs of previous nodes on the same path. So the optimal shortest path for the whole network topology is included in the partial topology the node knows. Using a shortest path algorithm, as Dijkstra’s algorithm [28], we can compute the optimal shortest path in the partial network topology. We can conclude that QOLSR3 finds the optimal shortest path. QOLSR3 finds the optimal shortest path using only partial knowledge of the network topology. The heuristic used in the QOLSR3 finds exactly optimal MPRs that guarantee minimum delay path. Nevertheless, there is no guarantee that this heuristic finds the minimum MPR set and also there is no optimization of the number of MPRs as done in Step 5 of QOLSR1 and QOLSR2 heuristics. Table 4 summarizes all proposed heuristics for the MPR selection comparing to the standard MPR selection of OLSR. In Section 3, we simulate three proposed QOLSR selection heuristics using the delay metric and
3150
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
Table 4 Proposed MPRs heuristics Heuristic
First metric
Second metric
Minimization of the MPR set
Standard MPR selection QOLSR1 QOLSR2 QOLSR3
Degree
None
Yes
Degree Delay Delay
Delay Degree Degree
Yes Yes No
compare them with the OLSR protocol. In following section, we present the routing table calculation performed by QOLSR. 2.6. Routing table calculation In traditional data networks, routing protocols usually characterize the network with a single metric such as hop-count or delay, and use the shortest path algorithms for path computation. For the delay metric, each arc (i, j) in the path p where p = (i, j, k, . . . ,q, r) is assigned a real number delayij. When the arc (i, j) is nonexistent, then delayij = 1. Let delay(p) = delayij + delayjk + + delayqr. The routing problem is to find a path p* between i and r so that delay(p*) is the minimum. In such a case, we use Dijkstra’s shortest path algorithm. 2.7. Path oscillation problem The consistent delivery of QoS guarantees requires stability of the data path. In particular, while it is possible that after a path is first selected, network conditions change and result in the appearance of new ‘‘better’’ paths, such changes should be prevented from unnecessarily affecting existing paths. In particular, switching over to a new (and better) path should be limited to specific conditions, e.g., when the initial selection turns out to be inadequate or extremely ‘‘expensive’’. This aspect is beyond the scope of QoS routing and belongs to the realm of path management, which is outside the main focus of this paper. However, because of its potentially significant impact on the usefulness of QoS routing, we briefly outline a possible approach to path management, according to [31]. In order to reduce oscillations between paths, we introduce a threshold value. The basic idea is to trigger path selection only when there is a significant change in the value of metrics compared to last metric measurement. The notion of significance of a change can be based on an ‘‘absolute’’ scale or a
‘‘relative’’ one. According to [31], an absolute scale means partitioning the range of values that a metric can take into equivalence classes and triggering an update whenever the metric changes sufficiently to cross a class boundary. A hysteresis mechanism may be added to suppress updates when the metric value oscillates around a class boundary. OLSR defines a ‘‘Link Hysteresis’’ strategy describing quality of the link that mitigates oscillations among paths. A relative scale, on the other hand, triggers updates when the percentage change in the metric value exceeds a predefined threshold. Independent of whether a relative or an absolute change trigger mechanism is used, a periodic trigger constraint can also be added. This constraint can be in the form of a hold-down timer, which is used to force a minimum spacing between consecutive updates. Alternatively, a transmit timer can also be used to ensure the transmission of an update after a certain time has expired. Such a feature can be useful if link state updates advertising bandwidth changes are sent unreliably. The QOLSR path selection, proposed in this paper, uses a relative scale, setting the threshold value to a predefined percentage. The strategy is: the metric measurement is only taken into account if and only if such a measurement exceeds a predefined threshold. Such a strategy mitigates oscillations and instabilities, which are typically generated by frequent changes on selected routes. Nevertheless, since the path oscillation problem involves path management, it is not completely solved only using our proposal, which do not include a path management. 3. Performance analysis The performance of the QOLSR protocol is studied with simulations. QOLSR has been implemented with OPNET modeler [32]. It is particularly popular in the ad hoc networking community, and many protocols used in ad hoc networks have been implemented, including Wireless LAN MAC layer IEEE 802.11 protocol, and some routing protocols such as AODV, DSR and TORA. Our implementation includes the OLSR module over the Wireless Module and QOLSR functions. We developed the OLSR protocol as specified in RFC3626. The implementation is completely modular and designed in compliance with other MANET protocols specified for radio/wireless models. In this way, such a modularity guarantees an easy
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
evaluation among MANET protocols and increases the accuracy in results while we use standard OPNET models. The MAC layer and PHY layer of the OPNET wireless model are designed based on the IEEE 802.11 standard [33]. Although the MAC implementation is not complete, it supports most of IEEE 802.11 standard functionalities, such as the back-off, deference, RTS/CTS, NAV, frame exchange sequence, fragmentation, access point functionality, basic service set, lost frame retransmission, and duplicate frame detection, etc. Both DCF mode and PCF mode are supported. In addition, OPNET supports the 802.11b physical layer model [33]. Both PLCP preamble and PLCP header are implemented and the multirate support is provided. There are three choices for the physical layer configuration in IEEE 802.11 standard: frequency hopping, infra red, and direct sequence. In our simulation scenario, we have selected the Direct Sequence Spread Spectrum (DSSS) technology. The configuration of the mobile stations is mostly based on the default parameter values proposed for the DSSS system in IEEE 802.11b standard [34], like the length of RTS, CTS, ACK, MAC header, the slot time, the SIFS, the DIFS, the PLCP preamble, etc. To provide the higher rates, 8-chip complementary code keying (CCK) is employed as the modulation scheme. 3.1. Simulation: mobile network scenario The main parameters used in simulations are given in Table 5. Networks of 30 nodes are generated, where nodes roam in an area of 1500 m by 300 m. The mobility model used is the random waypoint mobility model defining the way users move in the simulated area [35]. Network mobility is varied by changing the maximal nodal speed vmax = 10 m/s. The speed value is varied between 0 and 10 m/s to model different mobility. In such a model, each node picks a random destination in the defined area, sample a speed value. Once the node arrives at its destination, it pauses for a time p = 100 s. The network load is varied by changing the inter-arrival time of packets according to Table 5 parameters. The duration of the simulation is 300 s. Network traffic is generated by a traffic generating source, where the source and the destination are chosen randomly. This traffic generating source generates packets with 64 bytes. The time between successive traffic generations varies following a specified value. Such a value is varied changing the packet inter-
3151
Table 5 Parameters used in the simulations Parameter
Value
Transmission rate (Mbps) QOLSR a Packet size (bytes) Inter-arrival time (s) Number of nodes Mobility model Network area (m · m) Default distance (m) Maximum distance (m)
11 Varying between 0.3 and 0.7 64 Varying between 1 s and 0.0625 s 30 Random waypoint mobility model 1500 m by 300 m 100 300
arrival time between 1 and 0.08333, i.e. sources generate packets at a rate varying between 1 and 360 packets/s, varying the network load dynamically. All nodes are traffic generated sources. In order to assess the improvement achieved in QoS support capability by incorporating our algorithms to the original OLSR protocol, we implement three different heuristics for the MPR selection functionality. The implemented heuristics considering the minimum delay path are QOLSR1, QOLSR2 and QOLSR3 as detailed in Section 2.5.2. Then, we compare simulation results from the original best-effort OLSR protocol and our three QOLSR variants. Simulation results presented are based on a single metric for the routing calculation, i.e. the minimum end-to-end delay. Nevertheless, we have designed the routing decision (i.e. the metric used for the routing calculation) as well as QOLSR heuristics as simulation parameters. Then, for each simulation we can specify the metric used and the QOLSR heuristic proposed. Other implementations have been done using multiple-metrics, i.e. bandwidth and delay, for the routing calculation in [27]. The reason that we choose the delay constraint in our simulations, as discussed in Section 2.4.2, to calculate the available bandwidth for each node is still a challenging task. We believe that the delay constraint reflects a more realistic measurement of the quality of the network. 3.2. Simulation results The next figures show the relative performance of the routing algorithms OLSR, QOLSR1, QOLSR2, and QOLSR3. For each simulated routing algorithm, the QOLSR parameter ‘‘a’’ was varied within the set [0.3, 0.4, 0.5, 0.6, 0.7]. Results for a = 0.8 and a = 0.9 are not shown because for these values the algorithms become unstable and strongly scenariodependent.
3152
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
Fig. 4 depicts the number of received packets/sec for each simulated protocol. From the figure we can see that under light traffic, i.e. 50 packets/s, the number of received packets is about the same for all cases. This occurs because under light traffic the delay measurements present no significant variation to induce the selection of a new path. Therefore, as the main difference between QOLSR and OLSR is the inclusion of the delay metric, there is no considerable difference between their performance for the case of light traffic. In the case of more than 100 packets/s, the advantages of the proposed QOLSR protocols become apparent. Note that the greater is a, the lower is the weight of the last measured delay, i.e. the instantaneous delay. In this case, the instability caused by the instantaneous delay has little influence in the network behavior. For heavy traffic, i.e. more than 350 packets/s, QOLSR3 with a = 0.7 delivers fewer received packets than QOLSR3 with a = 0.3. This can be explained because when a = 0.3, and under heavy traffic, the weight of the last delay measurement is large and therefore the protocol change routes very often. The increase in the traffic increases the end-to-end delay, giving rise to path instability. For heavy traffic, the best results were achieved in the case of QOLSR1 with a = 0.7, where the improvement when compared with standard OLSR is around 12%. Finally, as a benchmark, in the figure we also plot a line
representing the maximal values, i.e. the ideal values, where 100% of the transmitted packets are received with zero dropped packets. Fig. 5 shows the average number of dropped packets/s for each simulated protocol. The number of dropped packets/s is considerably decreased for all QOLSR variants when compared with standard OLSR. The best performance was achieved for QOLSR1 with a = 0.7, where the packet loss is 58% smaller than for OLSR. Table 6 summarizes the gains of the QOLSR variants with respect to standard OLSR. Fig. 6 presents the average packet delay for each protocol. At an arrival rate of under 200 packets/s, QOLSR1 and QOLSR2 achieved a smaller average delay when compared with OLSR. In the plot we zoom in on the range between 30 and 220 packets/ s which make more clear the gain of the proposed QOLSR protocols for that range. Under light traffic, the best results are achieved by QOLSR1, followed by QOLSR2. From the figure we can also see that the relative performance changes depending on the traffic condition (after 220 packets/s). We have demonstrated in Section 2.5.5 that QOLSR3 finds the optimal path with respect to the delay metric. However, in our simulations we find that QOLSR3 generates more instability in the paths because such a protocol reacts more to the traffic variation, increasing the total average delay. Moreover, the good performance achieved
400 OLSR Maximal QOLSR1 (alpha=0.7) QOLSR2 (alpha=0.7) QOLSR3 (alpha=0.3) QOLSR3 (alpha=0.7)
350
Received (packets/sec)
300
250
200
150
100
50
0 0
50
100
150
200
250
300
Transmitted (packets/sec)
Fig. 4. Comparison of the average of received packets/s.
350
400
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
3153
80 OLSR QOLSR1 (alpha=0.7) QOLSR2 (alpha=0.7) QOLSR3 (alpha=0.7)
70
Dropped (packets/sec)
60
50
40
30
20
10
0 0
50
100
150
200
250
300
350
400
Transmitted (packets/sec)
Fig. 5. Comparison of the number of dropped packets/s.
by OLSR in Fig. 6 can be explained by the fact that in OLSR the number of dropped packets is larger than in the QOLSR protocols, and that such dropped packets are not taken into account in the average delay calculation. Then, when a packet is not dropped and is delivered with a large delay, the total average delay is increased. However, by
Table 6 Traffic dropped: the gain achieved for each simulated QOLSR protocol comparing with OLSR Protocol
Gain achieved (%)
QOLSR1 QOLSR2 QOLSR3
(33.18–62.65) (26.68–40.77) (14.58–40.07)
0.22 OLSR QOLSR1 (alpha=0.6) QOLSR1 (alpha=0.7) QOLSR2 (alpha=0.4) QOLSR2 (alpha=0.7) QOLSR3 (alpha=0.5) QOLSR3 (alpha=0.6)
0.03 0.2 0.18
0.028 0.026 0.024
0.16
Delay (sec)
0.14
0.022 0.02 0.018
0.12 0.1
0.016 0.014 40
60
80
100
120
140
160
0.08 0.06 0.04 0.02 0 50
100
150
200
250
Transmitted (packets/sec)
Fig. 6. Comparison between three proposed QOLSR algorithms of the average packet delay.
3154
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156 100 OLSR QOLSR1 (alpha=0.7) QOLSR2 (alpha=0.4) QOLSR2 (alpha=0.7) QOLSR3 (alpha=0.3) QOLSR3 (alpha=0.5)
90
Cumulative probability (%)
80 70 60
50 45 40 35 30 25 20 15 10 5 0 0.01 0.012 0.014 0.016 0.018 0.02 0.022 0.024
50 40 30 20 10 0 0
0.2
0.4
0.6
0.8
1
1.2
1.4
Delay (s)
Fig. 7. CDF of packets delay.
the same fact, QOLSR increases the network reliability. In order to investigate in detail why the average delay is reduced, we consider the cumulative distribution function (CDF) of packet delays. Fig. 7 shows the CDF of packet delays for a traffic load of 300 packets/s. In the figure zoom in on the range of 0–50%. From the figure we can see that all QOLSR protocols present better results than the standard OLSR protocol. We note that QOLSR3 achieves the largest delay reduction, about 44% less than OLSR. In the range of 60–95% in the CDF, the smallest delays are achieved by QOLSR1 with a = 0.7. For more than 95%, the best results are achieved by QOLSR2. The advantage of the three QOLSR variants over the standard OLSR protocol becomes evident when traffic gets heavy. Specifically, QOLSR3 achieves the best result due to its inherent improvement in the load-balance over the whole network. However, this improved load distribution is achieved through the increase of the MPR set, reflected in a increase in the control traffic over the network. For the case of light traffic, the performance results for the three proposed schemes are very similar, since in this case the routes are more stable. However, we can say that QOLSR1 and QOLSR2 have slightly better performance under light traffic because they minimize the MPR set and consequently generate less traffic control than QOLSR3.
4. Conclusions This article presented a QoS-based routing protocol for mobile and wireless ad hoc networks. In order to include quality parameters in the routing information, QoS measurements were applied. Methods to calculate delay and bandwidth measurements were proposed. The delay metric is calculated between each node and its neighbors having direct and symmetric links. The bandwidth measurements are calculated using IEEE 802.11b as the medium access control protocol. However, the throughput is very instable due to network conditions. Accurate throughput calculation is still a challenging task. Therefore, we considered delay as a metric rather than bandwidth as a metric. The implications of routing metrics on path computation were examined and the rationale behind the selection of QoS metrics were discussed. Heuristics for multipoint relays selection were proposed. The heuristic used in standard OLSR finds a MPR set with minimal size. However, there is no guarantee that OLSR finds the optimal path considering QoS constraints. Thus, three variants that allow QOLSR to find the minimum delay path were proposed. In order to include quality requirements in the MPRs selection, and also in routing information, delay measurements are applied. We demonstrated that the QOLSR3 heuristic finds the
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
optimal shortest paths using only partial knowledge of the network topology. The performance of the proposed QOLSR variants were examined through computer simulations. The three QOLSR variants achieve better performance when compared with the standard OLSR protocol. Finally, the proposed work can be used to adapt protocol functionalities such as route selection. It would even be possible to support adaptive applications, such as multimedia, which are sensitive to network changing conditions. Acknowledgements This work is supported by CNPq. The authors would like to thank Richard Demo Souza for his revision of the text, and the editor and the anonymous reviewers for their contributions that enriched the final paper.
[14] [15] [16]
[17] [18]
[19]
[20] [21] [22]
References [23] [1] S.-Y. Ni, Y.-C. Tseng, Y.-S. Chen, J.-P. Sheu, The broadcast storm problem in a mobile ad hoc network, in: ACM/IEEE SIGCOMM, Seattle, 1999. [2] E. Belding-Royer, C. Toh, A review of current routing protocols for ad-hoc mobile wireless networks, IEEE Personal Communications (1999) 46–55. [3] T. Clausen, P. Jacquet, Optimized Link State Routing Protocol (OLSR), IETF, RFC 3626, October 2003. [4] M. Steenstrup, Routing in Communication Networks, Prentice Hall International Inc., Englewood Cliffs, 1995. [5] J. Moy, OSPF Version 2, IETF, RFC 2328, April 1998. [6] Mobile Ad-hoc Networks IETF working group, Available from: http://www.ietf.org. [7] A. Qayyum, L. Viennot, A. Laouiti, Multipoint relaying technique for flooding broadcast message in mobile wireless networks, in: HICSS: Hawaii International Conference on System Sciences, Hawaii, 2002. [8] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler, A. Qayyum, L. Viennot, Optimized link state routing protocol, in: IEEE INMIC, Pakistan, 2001. [9] J. deMeer, A. Eberlein, Towards an advanced qos architecture, in: Multimedia Internet Broadcasting, Springer-Verlag, 2001. [10] I. Chlamtac, M. Conti, J.-N. Liu, Mobile ad hoc networking: imperatives and challenges, Elsevier Ad Hoc Journal 1 (1) (2003) 13–64. [11] R. Ramanathan, M. Steenstrup, Hierarchically-organized, multihop mobile wireless networks for quality-of-service support, ACM/Baltzer Mobile Networks and Applications 3 (1) (1998). [12] G. Bianchi, A. Campbell, R. Liao, On utility-fair adaptive services in wireless networks, in: IEEE IWQOS 98, Napa Valley, CA, 1998. [13] M. Naghshineh, M. Willebeek-LeMair, End-to-end QoS provisioning in multimedia wireless/mobile networks using
[24]
[25]
[26]
[27]
[28] [29]
[30]
[31]
[32] [33] [34] [35]
3155
an adaptive framework, IEEE Communications Magazine 35 (11) (1997) 72–81. N. Nikaein, C. Bonnet, A glance at quality of service models in mobile ad hoc networks, in: DNAC 2002, Paris, 2002. T. Logsdon, The Navstar Global Positioning System, Van Nostrand Reinhold, New York, 1992. D.L. Mills, Internet time synchronization: The network time protocol, in: Zhonghua Yang, T. Anthony Marsland (Eds.), Global States and Time in Distributed Systems, IEEE Computer Society Press, 1994. K. Roemer, Time synchronization in ad hoc networks, in: ACM MobiHoc, Long Beach, 2001. W. Su, I. Akyildiz, Time-diffusion synchronization protocol for sensor networks, IEEE/ACM Transactions on Networking 13 (2) (2005) 384–397. A. Munaretto, M. Fonseca, K.A. Agha, G. Pujolle, Virtual time synchronization for multimedia ad hoc networks, in: IEEE VTC 2004-Fall, 2004, Los Angeles. V. Jacobson, M.J. Karels, Congestion avoidance and control, in: ACM SIGCOMM, Standford, 1988. J. Postel, Transmission Control Protocol, IETF, RFC 793, September 1981. M. Kazantzidis, M. Gerla, End-to-end versus explicit feedback measurement in 802.11 networks, in: IEEE Symposium on Computers and Communications, Taormina, 2002. P. Gupta, P.R. Kumar, The capacity of wireless networks, IEEE Transactions on Information Theory IT-46 (2) (2000) 388–404. M. Grossglauser, D. Tse, Mobility increases the capacity of adhoc wireless networks, IEEE/ACM Transactions on Networking 10 (4) (2002) 477–486. S.H. Shah, K. Chen, K. Nahrstedt, Available bandwidth estimation in IEEE 802.11-based wireless networks, in: 1st ISMA/CAIDA Workshop on Bandwidth Estimation, San Diego, 2003. M. Heusse, F. Rousseau, G. Berger-Sabbatel, A. Duda, Performance anomaly of 802.11b, in: IEEE INFOCOM, San Francisco, 2003. H. Badis, A. Munaretto, K.A. Agha, G. Pujolle, QoS for ad hoc networking based on multiple metrics: bandwidth and delay, in: The Fifth IFIP International Conference on Mobile and Wireless Communications Networks, Singapore, 2003. M.J. Atallah, Algorithms and Theory of Computation Handbook, CRC Press, 1999. Z. Wang, J. Crowcroft, Quality-of-service routing for supporting multimedia applications, IEEE Journal of Selected Areas in Communications 14 (7) (1996) 1228–1234. A. Qayyum, L. Viennot, A. Laouiti, An efficient technique for flooding in mobile wireless networks, Research Report 3898, INRIA, June 2000. G. Apostolopoulos, D. Williams, S. Kamat, R. Guerin, A. Orda, T. Przygienda, QoS routing mechanisms and OSPF extensions, IETF, RFC 2676, August 1999. OPNET Modeler, Available from: http://www.opnet.com. OPNET Technologies, Wireless LAN Model Description, November 2000. IEEE 802.11 Standard, November 1999. J. Broch, D.A. Maltz, D.B. Johnson, Y.-C. Hu, J. Jetcheva, A performance comparison of multi-hop wireless ad hoc network routing protocols, in: ACM/IEEE MobiCom, Dallas, 1998.
3156
A. Munaretto, M. Fonseca / Computer Networks 51 (2007) 3142–3156
Anelise Munaretto is currently Associate Professor at the Federal Technologic University of Parana´ (UTFPR), Curitiba-PR, Brazil. She was graduated from the Pontifical Catholic University of Parana´ (PUC-PR), Curitiba-PR, Brazil with a B.Sc. in computer engineering in 1994. She received the M.Sc. and Ph.D. degrees in computer networks from the University Pierre et Marie Curie (Paris VI), Paris, France, in 2001 and 2004, respectively. Her research interests include routing and quality of service in mesh/sensor/ad hoc networks, WiMax, and wireless LAN.
Mauro Fonseca is currently Associate Professor at the Pontifical Catholic University of Parana´ (PUC-PR), Curitiba-PR, Brazil. He received the B.Sc. degree in computer engineering from PUC-PR, in 1994 and gained the M.Sc. degree in networks and distributed systems from the the Federal Center of Technological Education of Parana´ (CEFET-PR), Curitiba-PR, Brazil, in 1997. He received the Ph.D. degree in computer science from the University Pierre et Marie Curie (Paris VI), France, in 2003. His research interests are in service management frameworks and architectures for networks and beyond.
