Slot allocation schemes for delay sensitive traﬃc …vidya/papers/iitm2.pdf1. Introduction Hybrid wireless mesh networks are heteroge-neous multihop wireless networks that have both

ARTICLE IN PRESS

Computer Networks xxx (2005) xxx–xxx

www.elsevier.com/locate/comnet

Slot allocation schemes for delay sensitive traffic supportin asynchronous wireless mesh networks

V. Vidhyashankar a, B.S. Manoj b, C. Siva Ram Murthy c,*

a Department of Computer Science, Cornell University, Ithaca, NY 14853, USAb Department of Electrical and Computer Engineering, University of California San Diego, San Diego, CA 92093, USA

c Department of Computer Science and Engineering, Indian Institute of Technology, Madras, Chennai 600036, India

Received 17 October 2004; received in revised form 15 July 2005; accepted 26 September 2005

Responsible Editor: E. Gregori

Abstract

Heterogeneous multihop wireless networks such as wireless mesh networks consist of a set of resource-constrainedmobile nodes that want to communicate with each other and a set of fixed relay nodes that are equipped with betterresources and help in relaying data to the mobile nodes. Supporting real-time traffic in wireless mesh networks is consid-ered as a challenging problem. In a synchronous or asynchronous slotted environment, the position of reservation slots atevery link of an end-to-end path influences the end-to-end delay. The major contributions of this paper are the following:(i) an on-demand QoS routing protocol for asynchronous single channel wireless mesh networks, (ii) three heuristics for theslot allocation and positioning process in such networks, and (iii) the adaptations of slot allocation and positioning strat-egies for taking advantage of the recovery capacity effect of batteries of the mobile nodes. The heuristics we propose are theEarly Fit Reservation (EFR), Minimum Bandwidth Reservation (MBR), and the Position-based Hybrid Reservation(PHR). The EFR heuristic reserves bandwidth at the first available slot on every link on a link-by-link basis in the forwardpath. The MBR heuristic allocates bandwidth on the links in the increasing order of bandwidth and the PHR heuristicassigns bandwidth for every link at a position which is proportional to its location in the path. Recent studies on chemicalbattery characteristics show that a pulsed current discharge can extend the battery life compared to a continuous discharge.We have also adapted the basic slot allocation strategies, that are aimed at maximizing system throughput, to take benefitof the pulsed discharge model. The adapted heuristics are designed to provide an extended battery life and hence reduce thenumber of battery recharges. Simulation studies show that EFR performs better in terms of delay characteristics, whereasPHR and MBR provide better system throughput in terms of call acceptance rate. The adapted heuristics are found to beperforming better in terms of the number of deaths of mobile nodes and the average number of path breaks.� 2005 Elsevier B.V. All rights reserved.

Keywords: Asynchronous ad hoc wireless networks; Time sensitive traffic; End-to-end reservation; Reservation slot positioning; Hybridwireless mesh networks

1389-1286/$ - see front matter � 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.comnet.2005.09.032

* Corresponding author. Tel.: +91 44 2257 4361.E-mail addresses: [email protected] (V. Vidhyashankar), [email protected] (B.S. Manoj), [email protected] (C. Siva Ram

Murthy).

mailto:[email protected]



2 V. Vidhyashankar et al. / Computer Networks xxx (2005) xxx–xxx

ARTICLE IN PRESS

1. Introduction

Hybrid wireless mesh networks are heteroge-neous multihop wireless networks that have bothfixed and mobile wireless relay nodes. The fixedrelay nodes have better resources such as processingcapability and battery power and they do not orig-inate or terminate user traffic. The mobile nodesnot only originate/terminate user traffic, also for-wards data on behalf of other nodes. In additionto the problems added by the mobility, these mobilenodes are highly constrained in terms of computingand energy storage. Each mobile node in a hybridwireless mesh network can communicate with neigh-boring nodes and they use multihop routing proto-cols along with wireless relaying to communicatewith the nodes that are beyond their transmissionrange. Some of the fixed nodes may have connectiv-ity to the external wired network and hence theymay act as gateways to the Internet. Hybrid wirelessnetworks have high potential to provide an econom-ical alternative to the existing wide area wirelesscommunication infrastructures. Supporting real-time traffic in an asynchronous hybrid wireless meshnetwork remains a tough task. Existing workassumes a TDMA or CDMA over TDMA modelsfor handling the hidden terminals. In such cases,synchronization of nodes proved to be very expen-sive. The major advantages of wireless mesh net-

1

2

PATH I

S

Mobile Node

Fixed Relay Node

Packet

Traversal of Route Request

Fig. 1. An example hybrid wirele

works are high data support, quick deployment,high scalability, high extendability and low costper bit transferred. Fig. 1 shows an example hybridwireless mesh network topology which also indicat-ing a path between two nodes S and D. The fixedrelay nodes merely forward data and can also serveas a gateway to the Internet.

Our work in this paper, involves development ofa QoS routing protocol and slot allocation and posi-tioning schemes for asynchronous hybrid wirelessmesh networks. Since the mobile nodes are highlyenergy constrained in comparison to the fixednodes, we extend the routing and slot allocationschemes to be more energy efficient by utilizing thepulsed battery discharge mechanism.

1.1. Why an asynchronous environment?

While a synchronous TDMA environment sim-plifies the channel allocation scheme in a multihopwireless network, it faces several challenges in ahybrid wireless mesh networks. We, in this paper,focus our solution on an asynchronous hybrid wire-less mesh environment compared to a synchronousTDMA environment because of the following issues.

• Cost of synchronization: Environments like theTDMA require synchronization which is expen-sive in terms of both battery power and band-

3

D

GatewayNode

Internet

ss mesh network topology.

V. Vidhyashankar et al. / Computer Networks xxx (2005) xxx–xxx 3

ARTICLE IN PRESS

width and hence not attractive for a network withhighly mobile nodes. This is because, in manypractical situations, a Global Positioning System(GPS) may not always be available. Hence syn-chronization is achieved through broadcast ofcontrol packets which leads to flooding andhence is resource-wise expensive. On the otherhand, asynchronous environments do not haveto synchronize the entire network explicitly to asingle clock.

• Frequent partitioning and merging: In an ad hocnetwork or hybrid wireless mesh environment,the network is subjected to frequent partitioningand merging, especially when the nodes arehighly mobile. Consider an example of two net-work partitions A and B which are synchronizedwith respect to the time instants T1 and T2,respectively. When A and B merge, then theentire network may have to re-synchronize.Apart from the synchronization overhead, somecalls are bound to be getting dropped due to col-lisions or synchronization differences during thesynchronization period.

We have modified the DSR protocol [1] to enableit to support QoS routing with bandwidth reserva-tion in asynchronous wireless mesh networks. TheQoS path setup process has three major steps, viz.,the Bandwidth Feasibility Test Phase, the BandwidthAllotment Phase, and the Bandwidth Reservation

Phase. During the Bandwidth Feasibility Test Phase,each node appends the bandwidth reservation infor-mation in the Route Request packets. During theBandwidth Allotment Phase, the destination noderuns a bandwidth allocation algorithm using the res-ervation information available in the Route Requestpacket. In the Bandwidth Reservation Phase, timeslots allotted by the algorithm for each link are com-municated to the participating nodes through theRoute Reply packets. Reservation of each link ismade using the ResvRTS-ResvCTS-ResvACKthree-way handshake as proposed in [5]. This entiremechanism occurs in an asynchronous environment.By asynchronous environment, we mean that theentire network is not synchronized to a global clockas required in a TDMA system, for synchronizingthe super-frame. Here a node uses relative timeinformation to convey the exact positions of the res-ervation slots to a neighbor node. This is similar tothe asynchronous MAC protocol proposed in [5].We have proposed three heuristics for the slot allo-cation and positioning algorithm and studied their

performance using simulations. The presence of het-erogeneous wireless nodes within a hybrid wirelessmesh network demands new solutions for both rout-ing and bandwidth allocation. This is because, theprotocol design can utilize the resource-rich fixedmobile nodes more often than the resource con-strained mobile nodes. Therefore, we extended therouting protocol and slot allocation schemes to ben-efit the highly energy constrained mobile nodes.

This paper is organized as follows. Section 2describes the related work in this area. Section 3briefs the existing protocols used in this work. Sec-tion 4 describes the slot allocation framework andthe heuristics. Section 5 presents results of the sim-ulation of our scheme while Section 6 summarizesthe paper.

2. Related work

To the best of our knowledge, no study has beendone till date on slot allocation in an asynchronousenvironment. There exist a few solutions for the syn-chronous TDMA based environments, some ofwhich are described in this section. Lin proposed atable-driven QoS routing mechanism in [2] and anon-demand call admission control scheme in [3].The authors of [2] assume a TDMA environmentwith all nodes synchronized to a global clock. Ineach slot, different spreading codes are used foradjacent links for improving bandwidth reuse andreducing the effect of hidden terminals. The timeframe consists of two distinct phases. The first isthe control phase in which the network controlfunctions can be performed in a distributed fashion.Each node takes turns to broadcast its informationto its neighbors in a specified time slot. After the endof the control phase comes the data phase. This isthe phase in which nodes vie with each other forsending data. The scheme in [2] reserves slots forreal-time traffic using a QoS extension of the DSDVprotocol. According to the protocol, slot informa-tion is stored in the routing table and periodicallyexchanged. This information is in the form of setsof free slots. The bandwidth reservation problemin a TDMA-based scheme has been shown to beequivalent to the satisfiability (SAT) problem whichis known to be NP-complete [6]. Hence a heuristichas been proposed in [2] for assigning slots to eachnode. The path bandwidth for each of the nodes inthe routing table is calculated. During call setup, theslot information is used to make call admission con-trol. Thus, using this slot information, the system


ARTICLE IN PRESS

can determine whether a call can be accepted at thebeginning of the connection request. The slotassignment is performed at each node by forwardingreservation packets along the path. This is done byobserving the common free slots in the adjacentnodes (this involves calculation of the path andthe link bandwidths). This protocol, however,suffers from a high blocking probability.

The work in [3] proposes an application of thisalgorithm in an on-demand routing protocol.According to the protocol in [3], the path band-width is calculated at each node in the path whenit receives the Route Request packet. It performs afeasibility test at every node in the path bycomparing the status of its slots with the slot listthat is present in the Route Request packet. Ifthe node is an intermediate node and it passesthe feasibility test, it appends its free slots to theRoute Request packet and broadcasts it again.However, if it is the destination node and it haspassed the feasibility test, then it sends a Route

Reply packet back through the path. The nodesthen reserve the slots when they receive the Route

Reply packets.The work in [4] describes a bandwidth allocation

algorithm performed at the destination node. Thescheme is implemented with an on-demand routingprotocol. The source node broadcasts the RouteRequest packets. Each node that receives the Route

Request packet appends its reservation informationin the packet and broadcasts it. This goes on till theRoute Request packet reaches the destination node.At the destination node, a bandwidth allocationalgorithm allocates the slots to the links and sendsthem back through the Route Reply packets. Theallocation algorithm selects the link with the leastcurrent bandwidth (i.e., the bandwidth reserved sofar by the algorithm in each link) at every iteration.The protocol is proposed for a CDMA-over-TDMA based synchronous environment. The algo-rithm in [4] suffers from the way the links are chosenat every iteration of the algorithm, i.e., when linkshave the same minimum current bandwidth the tieis broken randomly. This may not produce goodresults all the time.

All these protocols, as mentioned before, did nothave the need to address the issue of the hidden ter-minal problem as it had already been taken care ofby the hardware mechanism. Moreover, the under-lying assumption of synchronization exerts pressureon the scarce resources like bandwidth and batterypower.

3. Reference protocols and models used in our work

In this section, we discuss the MAC and the rout-ing protocols used in the implementation of the slotallocation strategies. The algorithms proposed inthis work are applicable to both synchronousTDMA-based and asynchronous TDMA-based adhoc wireless networks, wireless mesh networks,and hybrid wireless mesh networks. In order to pro-vide the feasibility check and reservation of thebandwidth in the asynchronous environment dis-cussed above, we require a MAC protocol whichhas the respective capabilities. This is the reasonfor selecting the Real Time MAC (RTMAC) proto-col proposed in [5], a bandwidth efficient andasynchronous MAC protocol that supports bothreal-time (RT) and best-effort (BE) traffic.

3.1. The RTMAC protocol

In RTMAC, bandwidth is provided by dividingthe transmission time into successive super-frames.The super-frame is implemented as a reservationtable maintaining the time intervals during whichnodes within its transmission range (including itself)are busy. The bandwidth reservations for RT trafficare made by reserving variable-length time slots(conn-slots) on the super-frames. The essential con-cept in this approach is the flexibility of slot place-ment in the super-frame. A set of consecutivereservation slots (each of which has a duration oftwice the maximum propagation delay) is reservedby the pair of nodes involved in the RT session.The slot allocation differs from the TDMA schemesince no time synchronization is necessary forRTMAC [5] and all reservations are done withrespect to the relative time period by which theRT session starts. This is illustrated in Fig. 2.

The reservation process shown in Fig. 2 is athree-way handshake. The ResvRTS is initiated bythe sender, the ResvCTS is a means of acceptanceof the suggested conn-slot by the receiver and anindication that the destination node has reservedthe conn-slot while the ResvACK is needed to indi-cate that the bandwidth has been reserved at thesender�s side. The hidden terminal effect is alleviatedusing the three-way handshake protocol as neigh-bors of the two nodes reserve the time interval ofthe RT session in their super-frame when theyobtain the ResvACK or ResvCTS (as the casemay be). Whenever the conn-slot suggested by theResvRTS cannot be accepted by the MAC level des-

Fig. 2. Reservation mechanism used in RTMAC.


ARTICLE IN PRESS

tination node, it searches for a feasible conn-slot tothe sender through a ResvNCTS packet. The senderchecks whether the conn-slot suggested is feasible toit. If not, the reservation fails at the MAC layer.

3.2. The DSR protocol

In our work, QoS routing has been implementedas an extension of the DSR protocol [1]. We havemodified the protocol by piggybacking on the RouteRequest packets, the reservation information alongwith the relative time information for calculatingthe path bandwidth. According to the DSR proto-col, the Route Request packets are initiated fromthe source node and each of those nodes whichreceives the Route Request packet checks if it isthe destination node that the request has beenaddressed for. Otherwise, the Route Request packetis forwarded. The node�s address is appended to thefield indicating the traversed path in the Route

Request packet. When a Route Request packetreaches the destination node, it responds back witha Route Reply packet which travels along the tra-versed path as indicated by the Route Requestpacket. We assume the existence of bidirectionallinks in the network. When the Route Reply packetreaches the source node, it then starts sending datathrough the route that was suggested. During pathbreaks, Route Error packets are sent on both thesides of the broken link till they reach the senderand the destination.

3.3. Power management model

Different battery models have been used to char-acterize the batteries of mobile nodes. Typical bat-tery characteristics are specified in [8]. There arechiefly two models of battery discharge, viz., con-stant current discharge and pulsed current discharge[11]. The constant current discharge behavior ofstorage cells shows that the discharge curve is flatover a large time period and a gradual slope isdeveloped as the voltage reaches the cut-off voltage(Vcut) specified by its manufacturer. In a pulsed cur-rent discharge, the battery is made to switchbetween short discharge periods and idle periods(rest periods). Fig. 3 shows the performance of thebipolar lead-acid battery subjected to pulsed dis-charge current of six current pulses obtained bythe authors of [11]. While Fig. 3(a) shows the char-acteristics of the current density of the cell, Fig. 3(b)depicts the voltage characteristics for the pulsed dis-charge. After each discharge, which lasts for 3 ms,the cell is kept idle for 22 ms during which no dis-charging is allowed to take place. The cell is ableto recover and revert back to its initial open circuitvoltage during the first three rest periods. After thefourth current pulse the rest period of 22 ms turnsout to be inadequate for the full recovery of the cell.

We aim to reap the benefits of the pulsed dis-charge model by strategically placing reservationsso as to (possibly) enable recovery during the result-ing idle periods. An exponentially decreasing func-

025 50 10075 125 150

0.25

0.5

0.75

1.0

1.25

1.5

1.75

2.0

2.25

025 50 10075 125 150

2

4

8

10

12

Time (ms) Time (ms)

i (A

/cm

2 ) E (

v)i : Current density (Ampere/cm2)

E : Voltage (volts)

(a) (b)

Fig. 3. Pulsed discharge characteristics of Lithium-Ion cell.


ARTICLE IN PRESS

tion [11,12] depending on the current state (the dis-charge potential and the theoretical capacity) is usedto model the recovery. Thus, we have adopted arecovery model for the mobile nodes by whichthe battery may recover when the node is idle. Thebattery is said to be in a dormant state either whenits battery voltage has reached Vcut or when thetheoretical capacity has been exhausted.

4. Our slot allocation framework

The wireless mesh network has been imple-mented in an asynchronous environment withQoS-DSR as the routing protocol. The RTMACprotocol [5] is used for effecting the reservations inthe asynchronous environment. The QoS routingprotocol has three phases of operation, viz., theBandwidth Feasibility Test Phase, the Bandwidth

Allocation Phase, and the Bandwidth ReservationPhase.

• The Bandwidth Feasibility Test Phase: In thisphase, the selection of paths with the requiredbandwidth is performed. This is achieved duringthe Route Request propagation. Whenever a par-ticular node receives a Route Request packet, itchecks for bandwidth availability in the linkthrough which it is received. If sufficient band-width is available, then the Route Request is for-warded, hence avoiding the paths which cannotsupport the bandwidth requirement. Before for-

warding, the node synchronizes its reservationinformation with the already present informationin the Route Request packet that it received. Thisis piggybacked on the new Route Request packetwhich is then broadcasted.

• The Bandwidth Allocation Phase: During thisphase, the destination node performs a link-wiseconn-slot (the time interval needed for a singleRT session) allocation algorithm to assign theconn-slots at every link. The allotted conn-slot

information is included in the Route Reply packetand sent to the source node using the pathtraversed by the Route Request packet.

• The Bandwidth Reservation Phase: In this phase,the reservation of bandwidth in the form ofconn-slots at every intermediate link is carriedout. If the allotted conn-slot is not currently freethen any other free conn-slot can be reserved bythe MAC layer. The details of the reservationfailure is discussed in detail later in this section.This unavailability of the allotted conn-slot canbe caused by the simultaneous allotment of thesame conn-slot for a different flow during the per-iod between the feasibility check and the actualreservation.

These phases will be discussed in detail later inthis section. The mobile nodes in the mesh networkhave finite battery power. Since transmissioninvolves maximum power consumption, we assumethat the battery discharges only during transmis-


ARTICLE IN PRESS

sion. The battery of the node may experience acharge recovery [11] when the node is idle andrecharges whenever it is discharged fully. Once itis fully discharged, we assume that there is a shorttime period during which the node is unable to oper-ate. The node is then said to have attained a dor-

mant state or a death is said to have occurred. Thefixed relay nodes, on the other hand are assumedto have infinite power as they have a supportiveinfrastructure.

4.1. The QoS-DSR Protocol

In this section, we present a detailed discussionon the routing protocol and the slot allocationschemes.

1. The Bandwidth Feasibility Test Phase: Duringthis phase, the Route Request packets containingthe reservation information are forwarded till thedestination node is reached. Each node sends its res-ervation table along with the reservation frames ofthe links of the path covered so far. The reservationtable of a node consists of the information about thereservations of its neighbors and itself. The Route

Request carries the information stored in the reser-vation table which is in the form of reservationframes containing time durations. Each of the reser-vation frames belongs to one of the following states.

• ASYNC_FREE: This state is acquired whenthere is no transmission in the medium withinthe range of the node during that particular timeduration.

• ASYNC_UNRESV: This corresponds to theunreservable state of the medium within therange of the node when it is involved in transmis-sion (either as a sender or a receiver) during thatparticular time duration.

• ASYNC_HT: This corresponds to the unreserv-able state of the medium within the range ofthe node when one or more of its neighbors areinvolved in transmission during that particulartime duration (HT in ASYNC_HT refers toHidden Terminal).

The Route Request packet transmitted by anintermediate node contains the reservation framesof the nodes of the path traversed so far, in additionto the reservation table of the intermediate node.Whenever a node receives a Route Request packet,it uses the reservation information of the transmitter(of the Route Request) and its own to obtain a res-

ervation frame for the link. Then, using the newreservation frame for the current link and the reser-vation frames of the previous two links (obtained byusing the reservation frames of the last three nodesin the path traversed so far), the node checkswhether a call can be admitted or not. If the callcan be admitted, the node appends the reservationframe of the current link to the existing array andalso inserts its reservation table to the new Route

Request packet. A Route Request packet is for-warded by a node only if it satisfies the feasibilitytest. Along with the reservation information of thetraversed nodes, the Route Request will additionallyhave information on whether the traversed nodesare mobile or fixed relay nodes represented by aboolean value. An intermediate node, before for-warding the Route Request packet, updates the res-ervation information contained in the Route

Request.The destination node waits for a certain amount

of time decided by a parameter called Route Request

Window and one of the Route Request packets col-lected during this time interval, is chosen for thenext phase, viz., the Bandwidth Allocation Phase.The choice is made by the destination node bychoosing the Route Request packet with the maxi-mum number of fixed relay nodes in the path. Atie is broken by choosing the path with a lesser num-ber of hops. If there is a tie again, the one with thegreater bandwidth is chosen. Other Route Request

packets that arrive beyond the Route Request Win-

dow are discarded. This path selection process isdepicted in the example topology shown in Fig. 4.The figure shows three paths traversed by Route

Request packets for establishment of a connectionbetween nodes S and D. The three paths have thesame number of fixed relay nodes, but Path I (S-1-2-3-D) is chosen as it has the least hop count [10].

2. The QoS Frame: The data structure used forthe bandwidth allocation algorithm is called theQoS frame. This data structure is constructed justbefore the destination node runs the algorithm pre-sented in Fig. 9. The reservation information ofeach link, provided in the Route Request packet, isused for this purpose. The definition of the datastructure is useful in calculating the time complexi-ties of the algorithms. It is a linear array of blocks,each block containing a time duration and the stateto which it belongs. The state can be one of thosetabulated in Fig. 5. The table has been constructedwith respect to the directed link A–B as shown inthe figure. The state of the link depends on whether

1

2

3

4 5

6

7

PATH I

PATH II

PATH III

S

8

9 10

11

Mobile Node

Fixed Relay Node

Packet

Traversal of Route Request

Fig. 4. A scheme for choosing the path.

Node B State of the linkNode A

Free Free

Unreservable

Hidden Terminal

Unreservable

Unreservable

Free

Free

Free

Free

Hidden Terminal

Hidden Terminal

Unreservable

Unreservable

Unreservable

Hidden Terminal

Hidden Terminal

Hidden Terminal

ASYNC_FREE_FREE

ASYNC_FREE_UNRESV

ASYNC_FREE_HT

ASYNC_UNRESV_UNRESV

ASYNC_UNRESV_FREE

ASYNC_UNRESV_HT

ASYNC_HT_FREE

ASYNC_HT_UNRESV

ASYNC_HT_HT

B

A

Fig. 5. States with respect to the directed link AB.


ARTICLE IN PRESS

each of the concerned nodes is Free (the medium isfree in its vicinity), Unreservable (when the mediumis used by the node either as a sender or a receiver)or is a Hidden Terminal (when the node serves as ahidden terminal to a transmission involving one ormore of its neighbors). It is to be noted that thestates ASYNC_UNRESV_FREE and ASYNC_FREE_UNRESV are not possible in an ideal case.But they can arise either due to the unreliability ofthe medium or due to the transient nature of thenetwork caused by the mobility of the nodes. Con-sider, for example, two nodes A and B trying toreserve the channel using the three-way handshake(at the MAC layer). Suppose, the ResvCTS sent

by node B does not reach its neighbor node C.Before the data transmission between node A andnode B starts, if node C tries to broadcast a Route

Request packet for establishing some other connec-tion, the reservation information stored in the RouteRequest packet by node C will not contain the reser-vation made by node B. When the Route Request

packet reaches node B, it will store its reservationinformation that includes the recent most reserva-tion with node A. This is one possible scenario inwhich the state ASYNC_FREE_UNRESV mayarise.

3. The Bandwidth Allocation Phase: During thisphase, the bandwidth allocation for the path is per-formed at the destination node. The input to thealgorithm is the set of reservation frames con-structed from the reservation information in theRoute Request packet. These frames correspond tothose of the links of the path. The basic underlyingprinciple in the algorithm is that the bandwidth of alink is reserved by considering its frame along withthe two levels of neighboring links on both the sidesof this link. This is done for every link. The algo-rithm tries to avoid overlapping of reservation inneighboring frames at the first and at the secondlevel. By doing this, we try to alleviate the hiddenterminal problem and allot the necessary bandwidthat the same time. The heuristics, in general, varywith respect to the selection of the link at every iter-ation of the algorithm. We note at this point, that

ASYNC_HT_UNRESV ASYNC_HT_HTASYNC_HT_FREE

ASYNC_FREE_UNRESV

ASYNC_FREE_HT

SECOND_NGHR_1

SECOND NGHR 1

ASYNC_UNRESV_FREE

NGHR_1

SECOND_NGHR_2

SECOND_NGHR_2

SECOND_NGHR_2

ASYNC_UNRESV_UNRESV

ASYNC_FREE_FREE

SECOND_NGHR_1

LINK

SECOND_NGHR_2

SECOND_NGHR_2

NGHR_1

NGHR_2

ASYNC_UNRESV_HT

NGHR_2

SECOND_NGHR_1

SECOND_NGHR_1

SECOND_NGHR_1,

SECOND_NGHR_2

Fig. 6. State diagram describing the UpdateState function.


ARTICLE IN PRESS

there is a maximum error equal to the propagationdelay whenever the reservation information of anode is synchronized with its neighbor. The maxi-mum value of propagation delay is 1 ls when thetransmission range is 300 m. This is much smallerthan the guard band interval that is included inthe bandwidth that is allocated. This guard band

Fig. 7. A sample execution of t

on both the sides of the conn-slot helps in cushion-ing the error incurred in synchronization. The algo-rithm returns the time durations allocated for eachlink if it is successful. If the algorithm is unsuccess-ful, it returns NULL.

The function Resv_Bandwidth presented in Fig. 9contains the generic bandwidth allocation algo-

he UpdateState function.

Fig. 8. Algorithm to obtain free bandwidth.


ARTICLE IN PRESS

rithm. Four invocations of the GetAvailable_Band-width_Info function which is presented in Fig. 8finally produces the frame ResultantFrame whichcontains the free time durations during which aconn-slot may be reserved.

The algorithm tries to avoid overlapping of reser-vation in neighboring frames at the first and at thesecond level. Steps 2 (e � h) determine the timedurations in which the conn-slot can be reserved.The Steps 1, 2(i), and 2(j).(iii) depend on the heuris-tic used.

The UpdateState function used in the algorithmshown in Fig. 9 can be understood from the statediagram in Fig. 6. The states correspond to thoseof each time instant in the QoS frame of each link.Hence, all states shown in the diagram qualify to beinitial states (however, the function cannot be calledwhen the state for the time instant is ASYNC_UN-RESV_UNRESV). The transitions are labeled, eachlabel referring to the nature of the link, viz., LINK(the link for which the reservation is being made),NGHR_1 and NGHR_2 (the first neighboringlinks), and SECOND_NGHR_1 and SEC-OND_NGHR_2 (the second neighboring links). A

state transition occurs only for those time instantsfor which the UpdateState function has been called,also depending on the current nature of the linkspecified as a parameter of the function. A sampleexecution of the UpdateState function is shown inFig. 7. Fig. 7(a) shows the frames before the statetransition while Fig. 7(b) shows them after the statetransition.

(i) Early Fit Reservation (EFR): The EFR heu-ristic shown in Fig. 10 assigns bandwidthlink-by-link starting from the sender to thereceiver. At every link, the EFR tries to allo-cate the first available free conn-slot after thereserved conn-slot on the previous link. Theiterations are performed starting from the firstlink (that involves the source node), then pro-ceeding with the successive links of the path.This heuristic tries to reduce the end-to-enddelay of data delivery.

(ii) Minimum Bandwidth-based Reservation

(MBR): The MBR heuristic shown inFig. 11 allocates bandwidth to the links inthe increasing order of free conn-slots, i.e.,

Fig. 9. The generic algorithm.


ARTICLE IN PRESS

the link with the least free bandwidth isconsidered first. At every link, MBR allo-cates the first free conn-slot available. At eachiteration, the frame that has the minimumreservable bandwidth will be taken into con-

Fig. 10. The EFR

sideration. Thus we are trying to improve theprobability with which the algorithm is suc-cessful. The reservation for each link isdone by traversing from the beginning of theframe.

heuristic.

Fig. 11. The MBR heuristic.


ARTICLE IN PRESS

(iii) Position-based Hybrid Reservation (PHR):

The PHR heuristic shown in Fig. 12 assignsbandwidth slot for every link the position ofwhich is proportional to the position of thelink in the path. The operation of this heuristicis similar to the previous one. The differencelies in the selection of the conn-slot at a partic-ular position for reservation at each link. Thelink with the least free bandwidth is selectedfor each iteration. The reservation slot is posi-

Fig. 12. The PH

tioned for each link depending on its positionin the path.

We observe that this heuristic tries to reap thebenefits of both the EFR and the MBR heuristics.Thus positioning the conn-slot depending on thevalue of posn (see Fig. 12) will result in the earlierlinks in the path reserve (with a greater probability)in the beginning of the frame and the later links inthe path, reserve towards the end. Hence this type

R heuristic.


ARTICLE IN PRESS

of reservation yields a lesser delay than the MBRheuristic.

To adapt these heuristics to hybrid wireless meshnetworks, we have modified them to the EFR-Mesh,MBR-Mesh, and the PHR-Mesh. These adaptedheuristics take into consideration the power-con-strained nature of the mobile relay nodes. The mainobjective behind the adapted heuristics is to allocatea free time slot (hole) in a link just before a conn-slotduring the bandwidth allocation algorithm, when-ever the sender node of the link is a mobile relaynode. In other words, when a conn-slot has to beallocated to a link, the nature of the sender nodeis verified. If it is a mobile relay node, a conn-slot

is assigned such that there is a hole just before it.This, artificially created bandwidth hole is termedrecovery-hole. On the other hand, if the sender nodeof the link is a fixed relay node, the recovery-hole isnot allocated. The recovery-hole can possibly usefulto the mobile node for its battery to recover. Onlythe sender node is checked since the power manage-ment has been addressed only with respect to trans-mission in our model. Hence, the adaptive heuristicstry to reduce the possibility of the node reaching adormant state at the cost of fragmenting the reserva-tion frame. An illustration of the inclusion of arecovery-hole in the bandwidth allocation strategywhen EFR-Mesh is used in comparison with theEFR strategy is shown in Fig. 13. It compares theNAVs of the node B after call establishment whenEFR and EFR-Mesh are executed.

4. The Bandwidth Reservation Phase: If the band-width allocation algorithm is successful in its reser-vation, the bandwidth allocation information willbe attached to the Route Reply packet and sent to

A B C

B > A

EFRMesh

EFR

NAVs of Node B

B > C

Reservations Established:

recovery_hole

Fig. 13. Recovery slot allocation in EFR and EFR-Mesh.

the previous node in the path. When a node receivesa Route Reply packet, it then checks whether it is inthe path mentioned in the packet. If so, it reads thetime allotted for the link between itself and the nodewhich had sent the Route Reply. The node thenchecks from its super-frame whether it is free duringthe time interval suggested. If so, the call setup takesplace at the MAC layer using the RTMAC. Accord-ingly there is a three-way handshake taking placebetween the two nodes. If this three-way handshakeis successful, then the call gets established. TheRoute Reply packet is then sent to the next node.This goes on till the entire path reservation iscompleted.

5. Handling Bandwidth Reservation Failure: Dueto dynamic nature of the network and the temporaloverlap of multiple connection setup phases, it islikely that, at a link, a bandwidth slot allotted dur-ing the Bandwidth Allocation Phase may be unavail-able during the Bandwidth Reservation Phase. Thisis referred to as the Reservation Failure and it canbe handled in the following two ways: (i) the end-to-end reservation failure handling and (ii) the localreservation failure handling. According to the end-to-end reservation failure handling mechanism, theintermediate node that detects the Reservation Fail-

ure stops proceeding with the Bandwidth ReservationPhase and initiate ResvFailure packets to the sourceand the destination of the connection. Upon recep-tion of the ResvFailure packet, nodes in the down-stream (nodes on the path from the node thatoriginate the ResvFailure packet to the destinationof the connection) release their reservations andcache entries made for that connection. Theupstream nodes (nodes on the path from the nodethat originate the ResvFailure packet to the sourceof the path) prepare for releasing any cache entriesmade for the connection. The source of the path willthen initiate a new reservation process. In the localreservation failure handling scheme, the node thatdetects a Reservation Failure can locally allotanother free slot instead of the originally reservedone. We, in this work focused more on the perfor-mance of the slot allocation and positioning schemeand therefore used the end-to-end reservationfailure handling mechanism.

5. Simulation results

The proposed QoS-DSR routing protocol andthe heuristics for reservation slot positioning andallocation for the asynchronous hybrid wireless net-


ARTICLE IN PRESS

work environment were simulated using GloMoSim[7]. The heuristics were compared under identicalenvironments and loads. We have used ConstantBit Rate (CBR) sessions which generate datagrampackets of size 216 bytes, every 60 ms, for bothRT (Real Time) and BE (Best Effort) connections.We have used the radio capture model and thefree-space propagation model throughout the simu-lation. Simulation time was taken as 600 s and CBRsessions of 200 s were generated randomly between0 and 500 s of the simulation. The physical layerbandwidth for the wireless network interface istaken as 1 Mbps. This value for the physical layerbandwidth is taken to focus the study at high trafficload situations. The CBR sessions were classifiedinto two classes, namely RT and BE sessions.Terrain range used for the simulations was1000 m · 1000 m. Transmission range of a nodewas taken to be 300 m. The simulated network has30 nodes which were uniformly distributed in theterrain area. Random way-point mobility model isused. As mentioned before, an exponential distribu-tion is used to model the recovery and the value ofVcut follows that of [9]. The discharge and therecharge rates are assumed to be constant and thelatter is assumed to be ten times the former. The firstexperiment is a performance comparison of theEFR, MBR, and PHR heuristics with increasingnetwork load. Fig. 14 compares the average end-to-end delay performance for each of the three heu-ristics. Mobility is fixed at 20 m/s during this exper-iment. The EFR gives the least delay while the MBRyields the maximum delay and the PHR scheme

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

0 5 10 15 20 25 30

Del

ay (

sec)

Number of connections (BE = 0) mobility = 20 m/s

Avg-RT-Delay (EFR)Avg-RT-Delay (MBR)Avg-RT-Delay (PHR)

Fig. 14. Average end-to-end delay vs. network load (mobil-ity = 20 m/s).

provides an intermediate level of performance. Thereason for the performance of EFR can be attrib-uted to the manner in which conn-slots are assigned.In the EFR, we choose the first available conn-slotson successive links from source to destination.Hence the conn-slots are assigned in an early-fitmanner. Hence it is bound to give the least delaywhen compared to other heuristics. We estimatedthe end-to-end delay jitter as the standard deviationof the end-to-end delay. The average jitter com-pared in Fig. 15 also shows a similar trend as thatof the end-to-end delay. The jitter performanceshowed an increasing trend with increase in thenumber of connections in the network. This isattributed to the increased congestion in the net-work and the high mobility of nodes. Mobility,which is set at 20 m/s for this experiment, is alsofound to be contributing to the end-to-end jitter.Fig. 16 compares the throughput achieved by thethree heuristics, measured in terms of packet deliv-ery ratio. Packet delivery ratio performance showsthat the MBR and the PHR schemes provide amuch greater throughput than the EFR scheme.

In our experiments, we also conducted simula-tion experiments to compare the performance ofEFR, PHR, and MBR to their variations, adaptedfor hybrid wireless mesh networks called EFR-Mesh, PHR-Mesh, and MBR-Mesh, respectively.An important parameter that has been introducedin this set of results is the number of deaths of amobile node. This value is equal to the number oftimes a node reaches the dormant state.

The first set of simulations for comparing EFRand EFR-Mesh has been carried out with increasing

0.0124

0.0126

0.0128

0.013

0.0132

0.0134

0.0136

0.0138

0 5 10 15 20 25 30

Ave

rage

jitte

r


Average Jitter (EFR)Average Jitter (MBR)Average Jitter (PHR)

Fig. 15. Average jitter vs. network load (mobility = 20 m/s).

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

0 5 10 15 20 25 30

Pac

ket d

eliv

ery

ratio


Total Packet_Delivery_Ratio (EFR)Total Packet_Delivery_Ratio (MBR)Total Packet_Delivery_Ratio (PHR)

Fig. 16. Packet delivery ratio vs. network load (mobility = 20m/s).

18.5

19

19.5

20

20.5

21

21.5

22

20 21 22 23 24 25 26 27 28 29

Num

ber

of d

eath

s

Number of connections (BE = 0)

Average Deaths (EFR)Average Deaths (EFRMesh)

Fig. 18. Average number of deaths vs. network load (mobil-ity = 20 m/s).


ARTICLE IN PRESS

RT traffic load without any BE traffic. Fig. 17 com-pares the average end-to-end RT delay betweenEFR and EFR-Mesh. The end-to-end RT delay per-formance graph clearly shows that the EFR-Meshsuffers due to the allocation of recovery-holes. Thisis mainly because, the allocation of recovery-holes

resulting in an increased delay, that is proportionalto the length of the recovery-hole, at every interme-diate mobile relay node. Fig. 18 compares the aver-age number of deaths per node between EFR andEFR-Mesh schemes. The EFR-Mesh heuristic per-forms better than EFR heuristic owing to the recov-ery-hole allocated for every mobile relay node inorder to enable their batteries to recover the charge.

0.036

0.0365

0.037

0.0375

0.038

0.0385

0.039

0.0395

0.04

0.0405

0.041

20 21 22 23 24 25 26 27 28 29

Del

ay (

sec)


Avg-RT-Delay (EFR)Avg RT Delay (EFRMesh)

Fig. 17. Average end-to-end delay vs. network load (mobil-ity = 20 m/s).

Until now we discussed experiments with nobackground BE traffic and in order to study the per-formance behavior of the slot allocation and posi-tioning heuristics with the presence of BE traffic,we carried out more experiments as describedfurther in this section.

We have simulated the heuristics under a back-ground BE traffic load of 10 BE calls. Fig. 19 com-pares the average RT end-to-end delay experiencedby PHR and PHR-Mesh when background BE traf-fic is present. We noticed a moderate increase ofabout 10% in the delay of both PHR and PHR-Mesh compared to the case where no BE traffic ispresent. The delay difference between PHR and

0.058

0.059

0.06

0.061

0.062

0.063

0.064

20 21 22 23 24 25 26 27

Del

ay (

sec)


Avg-RT-Delay (PHR)Avg RT Delay (PHRMesh)

Fig. 19. Average RT end-to-end delay vs. network load (mobil-ity = 20 m/s).

0.075

0.08

0.085

0.09

0.095

0.1

0.105

0.11

20 21 22 23 24 25 26 27

Del

ay (

sec)


Avg-BE-Delay (PHR)Avg-BE-Delay (PHRMesh)

Fig. 20. Average BE end-to-end delay vs. network load (mobil-ity = 20 m/s).

3.8

4

4.2

4.4

4.6

4.8

5

5.2

5.4

5.6

20 21 22 23 24 25 26 27

Num

ber

of d

eath

s


Average Deaths (PHR)Average Deaths (PHRMesh)

Fig. 21. Average number of deaths vs. network load (mobil-ity = 20 m/s).

0.06

0.07

0.08

0.09

0.1

0.11

0.12

0.13

0.14

0.15

0.16

20 21 22 23 24 25 26 27

Cal

l blo

ckin

g ra

tio


CallBlockRatio (EFR)CallBlockRatio (EFRMesh)

Fig. 22. Average call blocking ratio vs. network load (mobil-ity = 20 m/s).


ARTICLE IN PRESS

PHR-Mesh remains almost constant. We alsostudied the variation of the end-to-end delay experi-enced by BE traffic flows when PHR or PHR-Meshheuristics are used for the RT traffic. It was interest-ing, in Fig. 20, to note that while the end-to-enddelay for RT traffic is almost remaining constantwith load, the end-to-end delay for BE traffic isfound to be increasing rapidly with number of RTconnections. This also shows that the bandwidthreservation for high priority RT traffic can indeedprovide a much higher quality of service in compar-ison to the BE traffic. In addition to the perfor-mance improvement for the RT traffic, our schemecan support both BE traffic and RT traffic simulta-neously. The call blocking ratios are comparedbetween EFR and EFR-Mesh with backgroundBE traffic load in Fig. 22. The graph shows that,on an average, the overall call blocking ratioincreasing rapidly when background traffic is pres-ent. This is because, the control packets for pathfinding and bandwidth reservation experiences highcontention from the background BE traffic. Also,we noted a slight increase of call blocking ratiofor the PHR-Mesh heuristics. This is mainly dueto the presence of fractured bandwidth in terms ofrecovery-holes that reduces the probability of callacceptance (see Fig. 21).

We now present the performance of the heuristicswith varying mobility. For this experiment, we keptthe number of RT and BE connections at 15 and 10,respectively. The mobility of nodes is varied from 2to 20 m/s. Each sample point is averaged over 10runs. We noticed that the trends for mobility are

the same for all the three heuristics. Fig. 23 showsthat the number of path breaks experienced per con-nection increases with mobility. This not onlyincreases the control packets but also influencesthe call dropping rate. The PHR-Mesh heuristicsperforms slightly better than the PHR heuristics.This is attributed to the fact that due to the fact thatthe call acceptance rate for PHR-Mesh is lower thanthe PHR heuristics. Fig. 24 shows that the call drop-ping ratios in PHR and PHR-Mesh increase withincreasing mobility, with the PHR-Mesh heuristicexhibiting slightly results. This increase in call drop-ping ratio is mainly attributed to the rapid change inthe topology where paths get broken quickly and

0.5

1

1.5

2

2.5

3

0 2 4 6 8 10 12 14 16 18 20

Num

ber

of p

ath

brea

ks

Mobility in m/s (RT = 15 BE = 10)

Average Path Breaks (PHR)Average Path Breaks (PHRMesh)

Fig. 23. Average number of path breaks vs. mobility.

0.15

0.2

0.25

0.3

0.35

0.4

0 2 4 6 8 10 12 14 16 18 20

Cal

l dro

ppin

g ra

tio


Total Dropping (PHR)Total Dropping (PHRMesh)

Fig. 24. Average call dropping ratio vs. mobility.

6

8

10

12

14

16

18

20

0 2 4 6 8 10 12 14 16 18 20

Num

ber

of d

eath

s


Average Deaths (EFR)Average Deaths (EFRMesh)

Fig. 25. Average number of deaths of mobile nodes vs. mobility.


ARTICLE IN PRESS

hence the system may not find enough bandwidthresources while attempting a path reconfiguration.Another reason for this increase in call droppingrate is the delay in releasing bandwidth upon detect-ing a broken path. The nodes in a path may takesome time to detect that the path is broken andhence the bandwidth reserved for the path remainsoccupied and unused for a short while. This is oneof the important contributors to the increase indropping rate. Again, due to the slight differencein the call acceptance rate in PHR-Mesh, it showsa slightly lower call dropping rate. We noted thatthe node mobility inversely affects the number ofdeaths as presented in Fig. 25 for EFR heuristics.This is found to be in line with the performance inpath breaks and call dropping rate. Since, the call

dropping rate and path breaks are increasing withmobility, the resultant number of active calls inthe system is less at high mobility. Therefore, thenumber of node deaths is found to be decreasingwith mobility. We also noted that the node mobilityalmost equally affects the three heuristics.

6. Conclusions

In this paper, we have proposed the following: (i)a QoS extension of the DSR protocol for hybridwireless mesh networks, (ii) three new slot allocationand positioning heuristics for supporting real-timetraffic in asynchronous multihop wireless networks,and (iii) adaptation of the proposed heuristics forthe asynchronous hybrid wireless mesh networks.The hybrid wireless mesh network is a multihopwireless network that consists of both fixed andmobile relay nodes. The QoS-DSR routing proto-col, proposed in this paper, is designed to providea stable route by preferring the fixed relay nodesover the mobile relay nodes. The routing decision,which is made at the destination node for selectinga path, is made based on the number of fixed relaynodes among the choice of available paths. The pro-posed new slot allocation and positioning heuristicsare Early Fit Reservation (EFR), Minimum Band-width-based Reservation (MBR), and Position-based Hybrid Reservation (PHR). The EFR heuris-tic attempts to allocate bandwidth, on links startingfrom the source to destination, choosing earliestavailable slot. MBR attempts to provide reservationon links starting from the minimum bandwidth link,in the order of increasing bandwidth, choosing the


ARTICLE IN PRESS

first available slot on each link. The PHR schemeprovides a novel mechanism in which the positionof the reservation slot, on a link, is placed in propor-tion to the link�s position on the path. We also haveadapted the slot allocation and positioning heuris-tics to support the highly resource constrainedmobile relay nodes by providing recovery-holes

along with bandwidth reservation, in order torecover the battery charge, in a hybrid wireless meshnetwork. We have carried out extensive simulationexperiments to evaluate the performance of the pro-posed heuristics. The EFR heuristic has been foundto give the best performance in terms of end-to-enddelay while the MBR provides better throughput.The PHR provides an intermediate performance incomparison with EFR and MBR. The heuristicsadapted to the hybrid wireless mesh environmentprovided an increased lifetime and fewer numberof node deaths for the power-constrained mobilenodes, with slight increase in the end-to-end delay.

References

[1] D.B. Johnson, D.A. Maltz, Dynamic source routing in adhoc wireless networks, in: T. Imielinski, H. Korth (Eds.),Mobile Computing, Kluwer, Boston, 1996, pp. 153–181.

[2] C.R. Lin, J.S. Liu, QoS routing in ad hoc wireless networks,IEEE Journal on Selected Areas in Communications 17 (8)(1999) 1426–1438.

[3] C.R. Lin, Admission control in time-slotted multihop mobilenetworks, IEEE Journal on Selected Areas in Communica-tions 19 (10) (2001) 1974–1983.

[4] H.C. Lin, P.C. Fung, Finding available bandwidth inmultihop mobile wireless networks. in: Proc. of IEEEVTC�00, vol. 2, May 2000, pp. 912–916.

[5] B.S. Manoj, C. Siva Ram Murthy, Real-time traffic supportfor ad hoc wireless networks, in: Proc. of IEEE ICON�02,August 2002, pp. 335–340.

[6] M.R. Garry, D.S. Johnson, Computers and Intractability,Freeman, San Francisco, CA, 1979.

[7] X. Zeng, R. Bagrodia,, M. Gerla, Glomosim: a library forparallel simulation of large-scale wireless networks, in: Proc.of IEEE PADS�98, May 1998.

[8] K. Lahiri, A. Raghunathan, S. Dey, D. Panigrahi, Battery-driven system design: a new frontier in low power design. in:Proc. of ASP-DAC/VLSI Design�02, January 2002, pp. 261–267.

[9] Panasonic Lithium Ion Batteries: Individual Data SheetCGR 17500. Available from: <http://www.panasonic.com/industrial/battery/oem/chem/lithion/>, April 2002.

[10] V. Vidhyashankar, B.S. Manoj, C. Siva Ram Murthy, Slotallocation strategies for delay sensitive traffic in multihopwireless networks, in: Proc. of HiPC�03, December 2003, pp.333–342.

[11] C.F. Chiasserini, R.R. Rao, Importance of a pulsed batterydischarge in portable radio devices, in: Proc. of ACMMOBICOM�99, August 1999, pp. 88–95.

[12] D. Panigrahi, C.F. Chiasserini, S. Dey, R.R. Rao, A.Raghunathan, K. Lahiri, Battery life estimation for mobileembedded systems, in: Proc. Int. Conf. VLSI Design�01,January 2001, pp. 55–63.

V. Vidhyashankar obtained his B.Tech. degree in Computer Sci-ence and Engineering in 2003 from the Indian Institute of Tech-nology (IIT), Madras, India. He is currently working towards thePh.D. degree in the department of Computer Science at theCornell University, Ithaca, USA.

B.S. Manoj completed his graduation in 1995 and post graduationin 1998 both in Electronics and Communication Engineeringfrom Institution of Engineers (India) and Pondicherry CentralUniversity, Pondicherry, India, respectively. He has worked as aSenior Engineer with Banyan Networks Pvt. Ltd., Chennai, Indiafrom 1998 to 2000 where his primary responsibility includeddesign and development of protocols for real-time traffic supportin data networks. He was an Infosys doctoral student during2000–2003 in the Department of Computer Science and Engi-neering at the Indian Institute of Technology (IIT) Madras, India,where he focused on the development of architectures and pro-tocols for Ad hoc wireless networks and next generation hybridwireless network architectures. During January 2004–January2005, he was a Project Officer at IIT, Madras, India. He is cur-rently a post-doctoral researcher at the University of California,San Diego, USA. Indian Science Congress Association hasawarded him the Young Scientist Award for the Year 2003. Hiscurrent research interests include Ad hoc wireless networks, nextgeneration wireless architectures, and wireless sensor networks.

C. Siva Ram Murthy received the B.Tech. degree in Electronicsand Communications Engineering from Regional EngineeringCollege (now National Institute of Technology), Warangal, India,in 1982, the M.Tech. degree in Computer Engineering from theIndian Institute of Technology (IIT), Kharagpur, India, in 1984,and the Ph.D. degree in Computer Science from the IndianInstitute of Science, Bangalore, India, in 1988.

He joined the Department of Computer Science and Engi-neering, IIT, Madras, as a Lecturer in September 1988, andbecame an Assistant Professor in August 1989 and an AssociateProfessor in May 1995. He has been a Professor with the samedepartment since September 2000. He has held visiting positionsat the German National Research Centre for Information Tech-nology (GMD), Bonn, Germany, the University of Stuttgart,Germany, the University of Freiburg, Germany, the SwissFederal Institute of Technology (EPFL), Switzerland, and theUniversity of Washington, Seattle, USA.

He is the co-author of the textbooks Parallel Computers:Architecture and Programming, (Prentice-Hall of India, NewDelhi, India), New Parallel Algorithms for Direct Solution ofLinear Equations, (John Wiley & Sons, Inc., New York, USA),ResourceManagement in Real-time Systems and Networks, (MITPress, Cambridge, Massachusetts, USA), WDM Optical Net-works: Concepts, Design, and Algorithms, (Prentice Hall, UpperSaddle River, New Jersey, USA), and Ad Hoc Wireless Networks:Architectures and Protocols, (Prentice Hall, Upper Saddle River,New Jersey, USA). His research interests include parallel anddistributed computing, real-time systems, lightwave networks, andwireless networks. He has published more than 100 internationaljournal and 100 international conference papers in these areas.

http://www.panasonic.com/industrial/battery/oem/chem/lithion/

http://www.panasonic.com/industrial/battery/oem/chem/lithion/


ARTICLE IN PRESS

Dr. Murthy is a recipient of Best Ph.D. Thesis Award from theIndian Institute of Science, Indian National Science Academy(INSA) Medal for Young Scientists, and Dr. Vikram SarabhaiResearch Award. He is a co-recipient of Best Paper Awards fromthe 5th IEEE International Workshop on Parallel and Distributed

Real-Time Systems (WPDRTS), and the 6th and 11th Interna-tional Conferences on High Performance Computing (HiPC).He is a Fellow of the Indian National Academy of Engineering

and an Associate Editor of IEEE Transactions on Computers.

Documents

Slot allocation schemes for delay sensitive traﬃc …vidya/papers/iitm2.pdf1. Introduction Hybrid wireless mesh networks are heteroge-neous multihop wireless networks that have both