Tsinghua Science and Technology Tsinghua Science and Technology

Volume 18 Issue 6 Article 1

2013

A Buffer Management Scheme for Packet Queues in MANET A Buffer Management Scheme for Packet Queues in MANET

Muhammad Aamir Department of Technology Support, National Bank, Karachi, Pakistan. This work was started when he was pursuing MS degree in information technology from SZABIST, Karachi, Pakistan

Mustafa A. Zaidi Department of Computing, SZABIST, Karachi, Pakistan

Follow this and additional works at: https://tsinghuauniversitypress.researchcommons.org/tsinghua-

science-and-technology

Part of the Computer Sciences Commons, and the Electrical and Computer Engineering Commons

Recommended Citation Recommended Citation Muhammad Aamir, Mustafa A. Zaidi. A Buffer Management Scheme for Packet Queues in MANET. Tsinghua Science and Technology 2013, 18(6): 543-553.

This Research Article is brought to you for free and open access by Tsinghua University Press: Journals Publishing. It has been accepted for inclusion in Tsinghua Science and Technology by an authorized editor of Tsinghua University Press: Journals Publishing.

TSINGHUA SCIENCE AND TECHNOLOGYISSNll1007-0214ll01/10llpp543-553Volume 18, Number 6, December 2013

A Buffer Management Scheme for Packet Queues in MANET

Muhammad Aamir� and Mustafa A. Zaidi

Abstract: We introduce a new scheme of buffer management to handle packet queues in Mobile Ad hoc Networks

(MANETs) for fixed and mobile nodes. In this scheme, we try to achieve efficient queuing in the buffer of a centrally

communicating MANET node through an active queue management strategy by assigning dynamic buffer space

to all neighboring nodes in proportion to the number of packets received from neighbors and hence controlling

packet drop probabilities. Our simulation study reveals that the proposed scheme is a way to improve the buffer

management for packet queues in MANET nodes in terms of packet loss ratio, transmission efficiency, and some

other important system parameters.

Key words: Active Queue Management (AQM); Mobile Ad hoc Network (MANET); queue management; packet drop

1 Introduction

Mobile Ad hoc Network (MANET) is a networkwith dynamic topology and mobile nodes. Due tothe dynamic nature of network, there is no centralcontrol. Hence, nodes communicate with other nodesthrough intermediate nodes. The intermediate nodes arenormal nodes in the same network and assume theresponsibility of forwarding packets on the route fromsource to destination.

When different services are the part of some network,it is necessary to give priority to the packets of delaysensitive services such as Voice over IP (VoIP) andvideo streaming applications. This is usually referredas Quality of Service (QoS) implementation in anetwork. When packets are processed according to theirassigned priority, more packets of the same or other

�Muhammad Aamir is with the Department of TechnologySupport, National Bank, Karachi, Pakistan. This work wasstarted when he was pursuing MS degree in informationtechnology from SZABIST, Karachi, Pakistan. E-mail:muhammad.aamir@nbp.com.pk; aamir.nbpit@yahoo.com.�Mustafa A. Zaidi is with the Department of

Computing, SZABIST, Karachi, Pakistan. E-mail:mustafainisb@gmail.com.�To whom correspondence should be addressed.

Manuscript received: 2013-06-21; revised: 2013-08-22;accepted: 2013-08-31

services also arrive usually at QoS processing hops. Inthis case, incoming packets are queued in buffer towait for their turn of processing. There are manyvariations introduced in the research of QoS paradigmabout how these queues are managed at processinghops. Moreover, the buffer size also plays an importantrole in terms of number of packets that can be held in aqueue before dropping the newly arrived packets (a caseof buffer overflow)[1].

The queue management scheme of Drop Tail hasbeen used for many years in which packets are droppedwhen the buffer is full. The length of buffer is thereforethe main parameter that controls the packet drop inthis scheme. Later, Active Queue Management (AQM)was introduced which is now prevailing in the networkworld. In this scheme, the sending node is notifiedbefore the queue is near to be completely filled sothat the sender can stop sending data or lower the rateof data transmission. Meanwhile, the current length ofqueue is shortened with the processing and de-queuingof buffered packets. After a sufficient space is againavailable in the queue, the source can be allowed to sendmore packets for en-queuing in the buffer and furtherprocessing[2].

The AQM is mainly used in wired networks thesedays where IPv4 packets are marked for their prioritiesin Type of Service (ToS) byte of the packet header. Nowfor the capable routers, Explicit Congestion Notification

544 Tsinghua Science and Technology, December 2013, 18(6): 543-553

(ECN) is sent to packet sending sources through twobits of the ToS byte. In this way, AQM is achievedin the networks’ QoS implementations[3]. On the otherhand, the mobile and wireless networks still use DropTail policies for packet queue management and theefforts to achieve AQM in those networks are verylimited. It is observed when AQM is implementedin the nodes of MANET, they still lack in efficientqueue management because packets are usually treatedfrom other wireless nodes only on the basis ofservices and applications. On the other hand, MANETnodes are resource constrained devices which havelimited battery life and memory/storage spaces. Insuch conditions, data sent from sources which transmitpackets with less data rates does not get a fair sharein queues. However, such sources are not capableto transmit packets with higher data rates due tocontention of nodes and interference areas on theroute. Taking it as an advantage, some other nodesbehave aggressively as they can send packets withhigher data rates to the destination node throughintermediate nodes. Therefore, they can occupy largerspace in the queues in unfair manner.

In this paper, we present a new scheme of buffermanagement for packet queues in mobile ad hocnetworks for fixed and mobile nodes. Using thisscheme, we try to achieve efficient queuing in thebuffer of a centrally communicating MANET node(we call it the queue management node or “QMN”)through an active queue management strategy byassigning dynamic buffer space to all neighboringnodes in proportion to the number of packets receivedfrom neighbors and hence controlling packet dropprobabilities. In the initialization phase, an equal bufferspace is allocated to all neighboring nodes whichenable every neighbor to have its share in the bufferof QMN in fair terms. In addition to that, as oursuggested algorithm is triggered on the occurrence of aselected incident, the allocation is dynamically adjustedaccording to instantaneous share of neighbors in theQMN’s buffer and the gap between the occupied andallocated buffer space. More importantly, we also putlimits on maximum and minimum buffer space a singlenode can occupy. In this way, a fair share in the bufferis available to all neighbors around a MANET nodeirrespective of packet transmission rates of neighboringsources. Since all neighbors share the buffer space, wecan mention that no node can occupy the buffer in fullthrough aggressively sending packets with higher data

rate. A neighbor has its share in the QMN’s buffer till itremains a neighbor of QMN.

2 Related Work

Kulkarni et al.[4] proposed and tested a queuemanagement scheme called PAQMAN againsttraditional RED algorithm in IP networks. Theirproposed scheme works on predictive measurement ofqueue size on the basis of traffic variations. The averagequeue length is predicted for next specified intervalusing recursive least square method and it determinesthe packet drop probability in that interval. Theirsimulation study in NS-2 depicts that the proposedscheme provides better QoS evaluations as comparedto RED. The same scheme is tested by Kulkarni etal.[5] on MANET environment and simulation study isperformed in NS-2 for packet loss and retransmissionefficiencies to check if the proposed scheme can workin an energy-efficient manner. The simulation resultsshow that PAQMAN coupled with ECN can reducepacket loss ratio and increase transmission efficiencywhile introducing negligible overhead. Friderikosand Aghvami[6] proposed “hop based queuing”which is an active queue management scheme forad hoc networks. In this technique, the packet dropprobability is proportional to number of hops a packettraverses. They indicate its advantages as reductionin mean excess delay on MANET and protection ofbuffer space from overflow due to such TCP flowswhich travel through many wireless hops and becomeunstable, i.e., the possibility of connection timeoutincreases. Lutz et al.[7] focused on the assignment oftransmission frames with same number of transmissionslots per frame to a wireless node on a channel sharedwith other nodes. They proposed a variation in such away that number of transmission slots (“weights”) canbe varied in different transmission frames. In this way,throughput may be increased without compromising“fairness” and packet losses due to collision mayalso be mitigated. Chen and Bensaou[8] presented astudy for high speed networks about their survivabilityin terms of fairness and packet loss problems withDrop Tail queue management scheme. The authorsmentioned when TCP flows come across multiplecongested links in high speed networks working onDrop Tail scheme, they face packet drop probabilityunfairness and round trip time unfairness. On the otherhand, AQM schemes reduce the severity of above

Muhammad Aamir et al.: A Buffer Management Scheme for Packet Queues in MANET 545

mentioned unfairness. Abbasov and Korukoglu[9]

improved the existing RED algorithm on networks andthe improvement is called Effective RED (ERED). Ithas a few variations as compared to RED in the packetdrop function which produce better throughput and lesspacket loss rate as compared to RED and some otherwell known AQM schemes. It is shown by authorsthrough simulation study in NS-2. Dimitriou andTsaoussidis[10] proposed an active queue managementscheme called Size-oriented Queue Management(SQM) in which the criterion is packet size. Hence, itdifferentiates time-sensitive traffic and applies differentpolicies of scheduling and packet drop on separateflows to increase the level of application satisfaction.

Our literature review identifies that considerablework has been done on the matter of packet queuemanagement in both wired and wireless forms ofnetworks. Some good efforts have also been made inthis direction for MANET environments. However, ahighly responsive solution is required that can addressthe packet drop issue of queue management in buffersof MANET nodes.

3 Proposed Approach

We propose a scheme of buffer management for packetqueues in MANETs for fixed and mobile nodes. For aMANET node, the packet queue is maintained in sucha way that an equal buffer space is allocated to eachneighboring source and an allowable extension is alsoavailable to each neighbor to avoid any underutilizationof resources. The allocation is made in the buffer of acentrally communicating MANET node and it is basedon number of packets received in the queue at node’sbuffer to utilize the buffer space efficiently withoutany monopolization of some surrounding source. Weconsider a MANET model working on Ad hoc On-demand Distance Vector (AODV) routing protocol. Itis a reactive protocol in which sources get routes todestinations when they demand for the same[11]. Nodesonly know their neighbors through routing table entriesand keep track of neighbors by exchanging HELLOpackets periodically.

We assume that we have a fixed node “QMN”surrounded by four neighbors “Node 1” to “Node 4”as shown in Fig. 1. There is another node, “Node 5”which is not treated as a neighbor of QMN at initialstage. The node QMN is analyzed in our model asa queue management node, i.e., it is used to allocate

Fig. 1 The considered MANET scenario.

buffer space to its neighbors according to our proposedscheme. We let the buffer space of QMN to be 100packets and it is considered empty at initial stage.

The node QMN allocates equal buffer space to all ofits neighbors in the initialization phase of allocation. If“bs” is buffer space and “nn” is number of neighbors ofQMN, then for each neighbor, the allocated buffer space“abs” is

abs Dbsnn

(1)

In our considered case, each of the four neighboringnodes gets the buffer space of 25 packets inQMN. However, it is necessary to consider if aneighboring node reaches its maximum allocation inQMN but the space is left in the total buffer due tothe reason that one or more other neighbors are notcommunicating or sending packets to QMN. Therefore,if we restrict a node to an allocation of 25 packets butthe space is still left in the buffer, there is a chancethat resource would be underutilized. So we provide anextension to each source which reaches or is about toreach the assigned limit of 25 packets upto a portion ofresidual buffer space, i.e., the instantaneous space leftin the buffer.

Here we assume that Node 1 has sent 25 packetsto QMN and reached the assigned limit in the bufferspace of QMN. However, a portion of the buffer spaceof QMN is still unoccupied because other neighbors arenot sending packets to QMN or they are communicatingwith lesser number of packets than their assigned limitof 25 packets per neighbor. Therefore, for an effectiveutilization of the remaining buffer space of QMN, wedetermine the residual buffer space “rbs” and divide itamongst all neighboring nodes in proportion to theirexisting occupied buffer spaces. This division is logicaland it effectively reduces the allocated buffer spacefor nodes which communicate with lesser numbersof packets but increases the same for nodes whichstand equal or closer to their assigned buffer space

546 Tsinghua Science and Technology, December 2013, 18(6): 543-553

limits. Therefore, we consider it fair at this stagebecause it gives a fair share in the buffer of QMNto nodes which communicate with higher numbers ofpackets. The extended buffer space “ebs” allocated toNode 1 is calculated through the relationship mentionedin Eq. (2).

ebsNode1 D

�rbsPnniD1 i

�� nn (2)

Here i represents the index of the number series from1 to number of neighbors. For better understanding, welet instantaneous occupied spaces in the buffer of QMNby the mentioned four neighbors be 25, 15, 22, and14 packets from Node 1 to Node 4, respectively. As25 packets is the maximum assigned limit at initialstage, it indicates that Node 1 has reached its limitwhereas total occupied buffer space is 76 packets byall neighboring sources, i.e., 24-packet buffer spaceis currently unoccupied. Therefore, the algorithm withour proposed scheme divides the residual buffer space,i.e., 24 with a summation figure obtained by adding anumber series from 1 to number of neighbors. In ourconsidered case, it is 4. Therefore, it divides 24 by10 (i.e., 4+3+2+1) and multiplies the result with thelargest number of the series which is 4 in our case. Thelargest number is multiplied so that the largest sharefrom residual buffer space can be allocated to the nodehaving reached the assigned maximum limit. The floorvalue is taken from the result and it is added to theinstantaneous buffer space occupied by Node 1. In ourconsidered case, 9 is added to 25 which extends themaximum buffer space allocated to Node 1 and it nowbecomes 34. Similarly, the neighbor that occupies thelargest buffer space in QMN after Node 1 is Node 3with 22 packets. It is about to reach its assigned limit of25 packets in QMN. Therefore, our proposed schemeshould provide it an extension in the buffer space. ForNode 3, the calculated ebs is

ebsNode3 D

�rbsPnniD1 i

�� .nn � 1/ (3)

From Eq. (3), we can check that 24 is divided by 10as before but the result is multiplied with the secondlargest number of the series which is 3 (i.e., 4�1)in our case. Here it shows that the series nn, nn�1,nn�2, � � � , 1 effectively contains normalization factorsalong with the summation of neighbors’ indices for theproposed buffer allocation scheme. The multiplicationfactor nn of Eq. (2) is replaced by nn�1 for Node3. The floor value is taken from the result and it isadded to the instantaneous buffer space occupied by

Node 3. In our considered case, 7 is added to 22which also extends the maximum buffer space allocatedto Node 3 and it now becomes 29. It shows that anextension is also provided to Node 3 in fair means asit is about to reach its maximum allocated space. Insimilar fashion, allocated buffer spaces for Node 2 andNode 4 are also adjusted and their new allocationsare 19 and 18 packets respectively. The multiplicationfactor nn of Eq. (2) is replaced by nn�2 for Node 2and nn�3 for Node 4. We can see that the maximumallowable buffer space is decreased for both nodesbut it is fair against adjusting the same for neighborscommunicating with higher numbers of packets. Afterall, we have to keep the total buffer space in QMNlimited to 100 packets (i.e., 34+19+29+18) accordingto our consideration. The rationale behind dividing rbswith a summation of neighbors’ indices and multiplyingwith components of the series of such indices is toassign the higher calculated limits to the neighboringnodes which are closer to assigned buffer space limitsin proportion to the number of packets received. Itcan be seen that the proposed scheme assigns dynamicbuffer space to neighboring nodes proportional to thenumber of packets received and hence controls packetdrop probabilities for them as discussed above.

An important consideration is how the algorithmrecalculates and maintains buffer space allocations forneighbors when packets are varied, i.e., increased ordecreased while processed in the buffer. First of all,we put limits on maximum and minimum buffer spacea single node can occupy. For this purpose, let usassume that a neighboring node cannot get bufferspace of more than 40 packets in QMN so that itcannot misuse the buffer through aggressive mode ofsending packets. Similarly, we put a lower limit of15 packets so that a node cannot be ignored in thebuffer space when it is a neighbor of QMN. Letus assume that the instantaneous occupied spaces inthe buffer of QMN by the mentioned four neighborsare now 27, 18, 25, and 18 packets from Node 1to Node 4, respectively. It means that Node 4 hasnow reached its last updated and assigned limit of18 packets whereas total occupied buffer space is88 packets by all neighbors, i.e., 12 packet bufferspace is now unoccupied. Since one of the nodes hasreached its assigned limit, the algorithm triggers again,recalculates buffer space allocations for neighbors asdiscussed above, and assigns corresponding limitsto neighbors. After recalculation for the mentioned

Muhammad Aamir et al.: A Buffer Management Scheme for Packet Queues in MANET 547

scenario, we get 30, 21, 27, and 22 packets as newlyassigned buffer space limits from Node 1 to Node4, respectively. The calculations are made and bufferallocations are adjusted only when a neighboring nodeapproaches the assigned upper limit of buffer spacein QMN so that the overhead of proposed schemeremains acceptable. In this way, dynamic buffer spaceis assigned to neighboring nodes proportional to thenumber of packets received and accordingly packetdrop probabilities are controlled. The working ofalgorithm is mentioned in Algorithm 1.

When number of neighbors is changed at an instant,i.e., a new neighbor appears and it is notified toQMN, the proposed scheme immediately reconfiguresassigned buffer spaces in such a way that equalshare is allocated to all neighbors including the newone. However, it does not affect the current occupancyof packets and the new neighbor has to wait forprocessing of packets in the queue until the spaceis available equal to its initially assigned limit inQMN. During this waiting period, the more incomingpackets from existing neighbors are dropped by QMNafter they reach their corresponding maximum limitsuntil the new neighbor holds its share in the bufferof QMN according to its initially assigned limit. Afterthat, the process continues as discussed before and thebuffer space is dynamically adjusted and allocated to allexisting neighbors according to instantaneous share ofneighbors in the QMN’s buffer and the gap between theoccupied and allocated buffer spaces. On the other hand,when a neighboring node is moved away and it is nomore a neighbor of the QMN then an additional portionof the buffer is available for existing neighbors. Inthis condition, the proposed scheme simply checks the

Algorithm 1 For each instance of a node approaching theassigned limit of buffer space1- Calculate total instantaneous buffer space occupied.2- Determine gap between assigned limit and buffer space occupied by eachnode.3- Arrange the gap values in ascending order with corresponding nodes.4- Obtain sum of number series (1! number of neighbors “nn”) say “Sum” .5- Obtain difference between total buffer space and total buffer spaceoccupied, i.e., residual buffer space “rbs”.6- Calculate (rbs / Sum) * nn and add it to buffer space occupied by the nodewith the least gap available through Step 3.7- Calculate (rbs / Sum) * (nn�1) and add it to buffer space occupied by thenode with the gap more and closer than the least available through Step 3.8- Repeat Step 7 for all remaining neighbors with decreasing value of “nn” by“1” each time and selecting nodes with respect to increasing values of gapsavailable through Step 3.9- Assign calculated buffer spaces to corresponding nodes as new buffer spaceallocations.

buffer space available against removal of the neighborand allocates equal share of the space to existingneighbors. In this way, maximum limit of currentlyallocated buffer space is effectively increased for eachexisting neighbor. For better understanding again, weassume that Node 5 mentioned in Fig. 1 has nowbecome a neighbor of QMN as shown in Fig. 2.

When QMN is notified about this neighbor afterreceiving HELLO message from Node 5, the proposedscheme reconfigures assigned buffer spaces in sucha way that equal share is allocated to all neighborsincluding Node 5. Let current allocated spaces fromNode 1 to Node 4 be 30, 21, 27, and 22 packets,respectively. Moreover, let current occupied spacesfrom Node 1 to Node 4 be 27, 18, 25, and 18packets, respectively. These values are considered incontinuation to the last updated status of buffer spacesdiscussed above. Now when Node 5 is also consideredas a neighbor of QMN for the first time, the bufferspace is divided into updated number of neighborswhich is now 5. A buffer space of 20 packets is nowto be assigned as the maximum allocated limit to eachneighboring node. However, as the current occupiedspaces are 27 and 25 packets in case of Node 1 andNode 3, respectively, Node 5 has to wait for processingof packets in the queue until space is available equal toits initially assigned limit of 20 packets in the QMN’sbuffer. During this waiting period, more incomingpackets from Node 1 and Node 3 are dropped until theyreach their new maximum limits of 20 packets eachand Node 5 holds its own share of 20 packets in thebuffer. At the same time, the buffer can accept 2 packetseach from Node 2 and Node 4 in the queue as their newallocated limits are 20 packets whereas they possess an

Fig. 2 The QMN node surrounded by five neighbors.

548 Tsinghua Science and Technology, December 2013, 18(6): 543-553

instantaneous space of 18 packets each. After that, theprocess continues as discussed before and buffer spaceis dynamically adjusted and allocated to all existingneighbors according to instantaneous share of neighborsin the QMN’s buffer and the gap between the occupiedand allocated buffer space. The algorithm is shown inAlgorithm 2.

We apply in the proposed scheme an active queuemanagement approach to notify a neighboring senderwhen its assigned limit is about to reach in the bufferof QMN. Upon receiving this notification, the neighborcan stop sending data or decrease the rate of sendingdata so that the current occupied space allocated tothat particular sender can be made sufficiently availableagain after processing and de-queuing of packets fromthe QMN’s buffer. After a sufficient space is againavailable in the queue for that particular neighbor, it canincrease the packet sending rate. The use of AQM in thisscheme is very useful to avoid packet losses becausea sender gets an alert to decrease sending packetson the expected occurrence of reaching its assignedbuffer space limit until a recalculated buffer space isprovided to accommodate more packets in the queuefrom that source. For this purpose, we apply ECN in thepacket header sent from QMN to its neighbor duringthe exchange of routing information to keep updatedrouting tables. In our analysis mentioned in the nextsection, we use a threshold value of 0.9, i.e., QMNsends ECN enabled packets to the neighbor when 90%of its allocated space in the buffer is occupied. When theneighbor is notified about congestion through this ECNinformation, it immediately responds by making itscongestion window one tenth of the existing width. Inthis way, early notification of congestion is madepossible through our AQM applied scheme whichminimizes packet losses and improves transmissionefficiency in the network.

It can be seen that the proposed scheme reflects theapproach of max-min fairness algorithm. The max-minfairness algorithm gives priority to data flows havingminimum flow rate. However, our proposed method

Algorithm 2 For each instance of appearance of a newneighbor1- Calculate total instantaneous buffer space occupied by each node.2- Increase neighbor count ‘nn’ by ‘1’.3- Divide total buffer space by ‘nn’.4- Assign calculated buffer spaces (equal space values obtained through step3) to all neighbors including the new one as updated buffer space allocations.

focuses on buffer space and gives priority to packetscoming from those nodes having more contributionin the communication as compared to others. Nowwe turn our attention towards the matters of trafficbursts and QoS requirements. The packet bursts can beobserved by QMN either from a genuine contributoror from some misbehaving node in the MANET. Theapplication of maximum buffer space limit is thefirst step towards protecting QMN’s buffer form burstof packets coming from a neighbor. Moreover, it isusually expected that a genuine neighbor will notalways or continually occupy the space in QMN’sbuffer equal to the maximum limit applied. If it isthe case with some neighbor, it is possibly be amisbehaving node and our scheme should have somesolution of this problem. Therefore, we design anobservation window with certain time intervals and analgorithm is programmed to observe nodes (neighbors)which occupy the maximum buffer space limit inQMN. If a neighbor is found having occupied maximumbuffer space limit for three consecutive observationintervals, we treat it as a misbehaving node for whichthe buffered packets are dropped and the assignedbuffer space limit of that certain neighbor is forcedto become the minimum, i.e., 15 packets in theQMN’s buffer. The recently available buffer space, thatbecomes free as a result of drop of the buffered packetsof misbehaving node, is distributed amongst remainingnodes (neighbors) in equal manner.

Even if the identified neighbor is not an intentionalmisbehaving node, the observation that it occupiesmaximum buffer space limit in QMN for threeconsecutive observation intervals gives an idea thatthis neighbor is becoming a cause of congestion,therefore the traffic drop of that particular nodeis reasonable. Moreover, a genuine neighbor wouldcertainly respond to ECN signals and stop sendingheavy traffic in a continuous mode. As a result, theprocessing of packets would leave more space availableto the node in QMN’s buffer and hence chances ofreaching the maximum limit are reduced. Even in thiscondition, if a neighbor is occupying the buffer spacewith maximum limit for three consecutive observationintervals then it must be a problematic node withmisbehavior or congestion. The algorithm is shown inAlgorithm 3.

We avoid any disconnection of the node fromMANET since this neighbor can be a genuine

Muhammad Aamir et al.: A Buffer Management Scheme for Packet Queues in MANET 549

Algorithm 3 For each instance of identification of amisbehaving node1- Drop buffered packets of misbehaving node (neighbor).2- Force the assigned buffer space of misbehaving node to become to theminimum limit to be assigned to a neighbor.3- Distribute the available buffer space (due to recent packet drops) amongstother neighbors in equal manner.4- Apply the updated buffer space allocations.

node trying to establish the communication withQMN. Therefore, it can build up its allocated bufferspace limit by communicating again with QMN and therest is done according to the designed scheme of bufferspace allocation. The sketch of observation window isgiven in Fig. 3.

It is shown in Fig. 3 that observation intervalsare named as tobs1; tobs2; and tobs3 for which thealgorithm waits before checking the occupancy ofnodes (neighbors) in the buffer of QMN. Intermediateintervals between tobs1 and tobs2, and tobs2 and tobs3

also exist during which the buffer occupancy of nodesis actually checked after an observation interval ispassed. These intermediate intervals also provide anextra space of time to neighbors so that they might geta chance to get rid of maximum buffer space occupancyin QMN (if observed) by making use of this marginof time. Each intermediate interval is not more than0.25 times of an observation interval. Moreover, theobservation intervals are equal in length, i.e., tobs1 D

tobs2 D tobs3: If a neighbor is found having occupiedmaximum allowable buffer space (determined as 40packets in our work) during tobs1; tobs2; and tobs3 alongwith two intermediate intervals in consecutive manner,the buffered packets of such a node is dropped from theQMN’s buffer in excess of 15-packet buffer space whichis the minimum configured limit of buffer space in ourwork.

The quality of service requirements of nodescan be treated in the considered model in a waythat the processing of packets in the buffer canbe implemented through any applied scheduling

Fig. 3 Sketch of the observation window to determinemisbehaving nodes.

mechanism of QoS. However, it applies on the packetde-queuing phase in the buffer. The proposed schemedeals with allocations of buffer space (which is actuallythe packet en-queuing phase in the buffer) and notwith the processing and priorities of packets. However,nodes with high QoS requirements normally tend todeliver more packets which increase node’s contributionin the communication and hence this scheme wouldautomatically offer more buffer space to such a node(neighbor). A point to clarify here is that we treat eachneighbor equally for packet en-queuing phase in thebuffer and allocate buffer spaces in accordance withnumbers of packets received. However, if a neighboris dealing with some delay sensitive applicationthen it is expected to send packets with markingsof priorities. Any implemented QoS mechanism (ifapplied) in QMN can then process the buffer accordingto the priorities of packets. In fact, the implementationof QoS mechanism is not an integral part of theproposed scheme, but if it is applied for the packet de-queuing phase then the overall processing of packetscan become capable to treat neighboring nodes in a QoSaware manner.

4 Performance Evaluation

We compare in this section the performance of ourproposed scheme with Drop Tail queue managementand another active queue management scheme withAODV routing scenarios created in the simulatorin terms of packet loss ratios and transmissionefficiencies. The active queue management schemeselected to be compared with our proposed schemeis ECN enabled PAQMAN[12]. We find that despitebeing an efficient scheme, a limitation of PAQMANis its predictive nature. It predicts the queue lengthin next prediction interval in terms of the average ofqueue lengths obtained in sampling intervals withinthe current prediction interval. However, the predictionmay not be accurate because a possibility always existsthat there can be a major difference in numbers ofpackets received in two consecutive intervals. In thiscase, the packet drop probability may give unexpectedresults by dropping higher numbers of packets or theinefficient utilization of resources. The responsivenessof PAQMAN is highly dependent on the value ofprediction interval. For shorter intervals, the schemeis more responsive but increases the computationaloverhead[4, 12]. On the other hand, the responsiveness

550 Tsinghua Science and Technology, December 2013, 18(6): 543-553

of our proposed scheme works with the current bufferutilization and the algorithm triggers when a senderreaches its assigned upper limit of buffer space andrecalculates buffer space allocations for neighbors.

For evaluation purpose, we consider a MANETwith 50 nodes in a hybrid environment of applicationusage. We first obtain the packet loss ratio in 50-node scenario where a particular node is selected asQMN in the unit of 5 nodes and other 4 nodes are itsneighbors. It means that we have 10 QMN nodes in ourscenario and each of them is surrounded by 4 differentneighbors. The applications we use are VoIP, FTP,Database, and Web services. The buffer space in eachnode is configured as 100 packets and network trafficmodel is Poisson distribution model. The packet lossratio is first obtained in 50-node scenario for differentflow arrival rates mentioned in Table 1.

The comparison in Table 1 indicates that ourproposed scheme works better in terms of packet lossratio as compared to Drop Tail and PAQMAN schemesfor tested flow arrival rates. On the flow arrival rate of25 Mbps, we observe that Drop Tail is slightly betterthan PAQMAN which indicates that PAQMAN maybe beaten by Drop Tail in low congestion cases. Wecan mention that as flow arrival rate increases, thereare more incidents of congestion due to more packettransmissions. Therefore, the packet loss ratio in allschemes generally increases with the rise in flow arrivalrate.

We now increase the number of nodes in MANETand tested two more scenarios with 150 and 250nodes. We make the same analysis to check theperformance of our proposed scheme in denseenvironments where chances of congestion increase.

The comparisons in Table 2 and Table 3 indicatethat our proposed scheme remains better in termsof packet loss ratio as compared to Drop Tail andPAQMAN schemes for tested flow arrival rates in thecase of scenarios with 150 and 250 nodes. In Figs. 4-

Table 1 Packet loss ratio in 50-node scenario.

Flow arrival rate Packet loss ratio(Mbps) Drop Tail PAQMAN Proposed

10 0.06 0.05 0.0325 0.23 0.24 0.1135 0.36 0.29 0.1945 0.41 0.33 0.2654 0.56 0.45 0.31

Table 2 Packet loss ratio in 150-node scenario.

Flow arrival rate Packet loss ratio(Mbps) Drop Tail PAQMAN Proposed

10 0.09 0.08 0.0625 0.31 0.28 0.1735 0.42 0.32 0.2245 0.54 0.38 0.2954 0.61 0.46 0.38

Table 3 Packet loss ratio in 250-node scenario.

Flow arrival rate Packet loss ratio(Mbps) Drop Tail PAQMAN Proposed

10 0.11 0.09 0.0725 0.34 0.30 0.2435 0.48 0.35 0.2945 0.59 0.46 0.3454 0.64 0.51 0.43

6, we present the observed transmission efficiencies ofconfigured network in tested scenarios of 50, 150, and250 MANET nodes.

We analyze in Figs. 4-6 the transmission efficienciesin MANET of our configured 50-node, 150-node,and 250-node scenarios to check the performance ofproposed scheme in light as well as dense environmentsof high congestion. The comparisons show thatour proposed scheme provides better transmissionefficiency in MANET as compared to Drop Tail andPAQMAN schemes for tested flow arrival rates. Wecan mention that with the increase of flow arrivalrate, more congestion occurs in the network becauseof more packet transmissions. Therefore, the overalltransmission efficiency in the network in all schemes

Fig. 4 Flow arrival rate vs. transmission efficiency in 50-node scneario.

Muhammad Aamir et al.: A Buffer Management Scheme for Packet Queues in MANET 551

Fig. 5 Flow arrival rate vs. transmission efficiency in 150-node scneario.

Fig. 6 Flow arrival rate vs. transmission efficiency in 250-node scneario.

generally decreases with the rise in flow arrival rate.We check in the next phase of simulation analysis,

the outcome of proposed scheme using QoS markingsof high priority VoIP packets. As mentioned before,the proposed scheme deals with allocations of bufferspace (which is actually the packet en-queuing phasein the buffer) and not with the processing and prioritiesof packets. Although the implementation of QoSmechanism is not an integral part of the proposedscheme, but if it is applied for the packet de-queuingphase then the overall processing of packets canbecome capable to treat neighbors in a QoS awaremanner. Therefore, in order to check the impact onsome QoS related system parameters, we mark packetsof different applications for their priorities underDiffServ implementation[1]. The VoIP packets are giventhe highest priority[13] and we check throughput, end-to-end delay, and jitter of the service as compared toother tested schemes, i.e., Drop Tail and PAQMAN. We

also increase the buffer space in each node upto 500packets as we generate heavy VoIP traffic with largenumbers of packets. Network traffic model is againPoisson distribution model.

We analyze in Figs. 7-9 the throughput, end-to-enddelay, and jitter respectively of VoIP application inMANET of our configured 50-node, 150-node, and250-node scenarios to check the impact of proposedmethod on some important system parameters underDiffServ implementation of QoS packet markings. Thecomparisons show that our proposed scheme, inconjunction with DiffServ implementation of QoSpacket markings, provides better throughput, end-to-end delay, and jitter of VoIP application in MANETunder configured settings as compared to Drop Tail and

Fig. 7 Throughput of voice over IP application.

Fig. 8 Packet end-to-end delay in voice over IP application.

Fig. 9 Packet end-to-end jitter in voice over IP application.

552 Tsinghua Science and Technology, December 2013, 18(6): 543-553

PAQMAN schemes for tested flow arrival rates. Wecan mention that with the increase of flow arrival rate,more congestion occurs in the network because of morepacket transmissions. Therefore, the overall throughputdecreases whereas end-to-end delay and jitter of VoIPapplication generally increase in all schemes with therise in flow arrival rate.

Finally, in order to estimate the overhead of ourproposed scheme, we first calculate total number ofcontrol packets of AODV used for communicationsamong the MANET nodes using our 50-node scenariowith simple Drop Tail scheme. Later, we calculate totalnumber of control packets of AODV routing protocoland control packets of our proposed scheme for thesame scenario of communications among the MANETnodes. We find that there is about 4% increase inthe total number of control packets in MANET dueto our proposed scheme. It is acceptable for largeMANETs. In fact, the overhead of proposed schemedepends on the frequency of packet sending ratesfrom neighbors as rapid changes in the buffer spaceallocations in QMN trigger the respective algorithmsrepeatedly in quick manner. Moreover, quick changesin numbers of neighbors around QMN also triggerthe algorithms frequently. This processing overheadintroduced by the proposed scheme is howeverovercome by the efficient and well-distributed bufferspace allocations during packet en-queuing phases andhence we obtain reduced losses and end-to-end delayswith better transmission efficiency. Therefore, we canmention that the proposed scheme is still an efficientbuffer management scheme to handle packet queuesin mobile ad hoc networks for improved queuing inMANET nodes.

5 Conclusions and Future Work

We present in this paper a new scheme of buffermanagement to handle packet queues in mobile adhoc networks for fixed and mobile nodes. In thisscheme, we obtain improved queuing in a MANETnode through an active queue management process byassigning the dynamic buffer space to all neighboringnodes proportional to the number of packets receivedand hence controlling packet drop probabilities. Inthe initialization phase, an equal buffer space isallocated to all neighboring nodes which enable everyneighbor to have its share in the buffer of the nodein fair terms. Later, when the proposed algorithm is

triggered on the occurrence of a selected incident,the allocation is dynamically adjusted according tothe instantaneous share of neighbors in the node’sbuffer and the gap between the occupied and allocatedbuffer space. We also put limits on maximum andminimum buffer space a single neighbor can occupyin the node’s buffer and a neighbor has its sharetill it remains a neighbor of the node. Simulationstudy indicates that the proposed scheme is a wayof getting improved buffer management for packetqueues in mobile ad hoc network nodes. We simulatedthis scheme for packet loss ratios and transmissionefficiencies in 50-node, 150-node, and 250-nodescenarios and compared its performance with DropTail and PAQMAN schemes. We also checked theperformance of proposed scheme in conjunction withDiffServ implementation of QoS packet markings forVoIP traffic in terms of throughput, packet end-to-end delay, and jitter statistics and found it better ascompared to Drop Tail and PAQMAN schemes. Infuture, we have a plan to further test this schemein MANETs having environments of higher numbersof nodes, high mobility, more applications, and morevariations of flow arrival rates. We also aim to reducethe overall processing overhead of proposed schemethrough modifications in suggested algorithms.

Acknowledgments

The authors are thankful to Mr. Kamran Ul Haque,Systems Development Wing, NBP, Karachi, Pakistan,for developing codes to execute the simulation workmentioned in this paper. We are also thankful toanonymous reviewers for their constructive comments andvaluable suggestions to improve the quality of this paper.

References

[1] M. Aamir, M. Zaidi, and H. Mansoor, Performanceanalysis of Diffserv based quality of service in amultimedia wired network and VPN effect using OPNET,International Journal of Computer Science Issues, vol. 9,no. 3, pp. 368-376, 2012.

[2] L. Chrost and A. Chydzinski, On the evaluation ofthe active queue management mechanisms, in Proc. 1stInt. Conf. On Evolving Internet (INTERNET), 2009,pp. 113-118.

[3] K. R. S. Floyd and D. Black, The addition of explicitcongestion notification (ECN) to IP, RFC 3168, Sep. 2001.

[4] P. G. Kulkarni, S. I. McClean, G. P. Parr, and M. M. Black,Proactive predictive queue management for improved QoSin IP networks, in Proc. IEEE ICN/ICONS/MCL 2006,2006.

Muhammad Aamir et al.: A Buffer Management Scheme for Packet Queues in MANET 553

[5] P. G. Kulkarni, M. Nazeeruddin, S. I. McClean,G. P. Parr, M. M. Black, B. W. Scotney, and P. Dini,Deploying lightweight queue management for improvingperformance of mobile ad-hoc networks (MANETs), inProc. Int. Conf. On Networking and Services (ICNS), 2006.

[6] V. Friderikos and A. H. Aghvami, Hop based queueing(HBQ): An active queue management technique for multihop wireless networks, in Proc. IEEE ICICS/PCM 2003,2003, pp. 1081-1085.

[7] J. Lutz, C. J. Colbourn, and V. R. Syrotiuk, Variable weightsequences for adaptive scheduled access in MANETs,in Sequences and Their Applications � SETA 2012,T. Helleseth, J. Jedwab, Ed. Springer Berlin Heidelberg,2012, pp. 53-64.

[8] S. Chen and B. Bensaou, Can high-speed networks survivewith DropTail queues management? Comput. Netw.,vol. 51, no. 7, pp. 1763-1776, 2007.

[9] B. Abbasov and S. Korukoglu, Effective RED: An

algorithm to improve RED’s performance by reducingpacket loss rate, J. Netw. Comput. Appl., vol. 32, no. 3,pp. 703-709, 2009.

[10] S. Dimitriou and V. Tsaoussidis, Promoting effectiveservice differentiation with size-oriented queuemanagement, Comput. Netw., vol. 54, no. 18, pp. 3360-3372, 2010.

[11] C. Perkins, E. Belding-Royer, and S. Das, Ad hoc on-demand distance vector (AODV) routing, RFC 3561, July2003.

[12] K. D. Kumar, I. Ramya, and M. R. Masillamani, Queuemanagement in mobile adhoc networks (Manets), inProc. IEEE GreenCom/CPSCom 2010, 2010.

[13] M. Aamir and M. Zaidi, QoS analysis of VoIP trafficfor different codecs and frame counts per packet inmultimedia environment using OPNET, in Proc. 15thIEEE Int. Multitopic Conf., 2012, pp. 275-281.

Muhammad Aamir received his BEngdegree in electronics from Institute ofIndustrial Electronics Engineering (IIEE),Karachi, Pakistan in 2006. He got hisMS degree in information technologyfrom Shaheed Zulfikar Ali Bhutto Instituteof Science & Technology (SZABIST),Karachi, Pakistan in 2013. He served Al-

Futtaim Technologies Pakistan (Pvt.) Ltd. as systems engineerfor deployment of ePABX, VoIP, and data communicationsolutions of Alcatel-Lucent at various sites. He is currentlyworking as technology support officer in NBP for differentapplication & networking domains. His current research interestsinclude computer networks, IP based communication systems,and network security. He holds the membership of IEEE.

Mustafa A. Zaidi is a full time researchscholar at Faculty of Computing, ShaheedZulfikar Ali Bhutto Institute of Science& Technology (SZABIST), Karachi,Pakistan. He received Master of AppliedPhysics degree from Karachi University in1994, Master of Power Engineering degreefrom NED University Karachi in 1997, and

Master of Computer Science degree from SZABIST in 2010. Hereceived the merit scholarship during Master degrees and iscurrently receiving PhD scholarship from Higher EducationCommission of Pakistan (HEC). His research interests includecomputer networks, communication systems and optimizationof multivariable quality characteristics using statistical andmachine learning tools.