Sie sind auf Seite 1von 5

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 10, OCTOBER 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 46

T-Drop: An optimal buffer management policy


to improve QOS in DTN routing protocols
Qaisar Ayub, Sulma Rashid

Abstract—Network architectures like TCPIP, AODV, OSPF perform communication in environments where end-to-end path must exist be-
fore the start of transmission and is not devisable in most of advanced wireless applications for example military networks , vehicular ad hoc
networks, pocket switched networks, deep space communication. Delay Tolerance network (DTN) build communication infrastructure
through intermittently connected mobile nodes where each node store the message in its buffer, carries the message while moving and for-
ward it when encounter with other nodes. To maximize the delivery probability, a node while moving replicate message copies to all encoun-
ters nodes. This iterative replication and storage of message produce congestion which is override by dropping message(s). An efficient buf-
fer management policy to improve quality of service is required to decide which message will be dropped, when buffer run out of its capaci-
ty.In this paper we proposed a new buffer management policy which drops the message from congested buffer only if size of existing queued
message(s) falls in Threashhold range (T).This technique is called as T-Drop. We prove through simulations that propose T-Drop policy mi-
nimize the message drop, hop count average, overhead while increase the delivery probability as compared to existing DOA.

Index Terms — Algorithms. Network communication, Routing protocols, Store and forward networks.

——————————  ——————————

1 INTRODUCTION

T he networking architecture (TCP/IP), Ad hoc [1]


works under the assumption that end-to-end path
between source and destination must exist before
mance of router. A critical issue is to select which mes-
sage(s) to be dropped upon a full buffer, therefore effi-
cient drop policy is necessary to overcome congestion
start of transmission and thus not applicable in most of amd to improve quality of service.
emerging wireless applications like military networks[2] , Some recent work [12],[13],[14], [15] addressed few
vehicular ad hoc networks (VANETs) [3], pocket switch- transmission scheduling and buffer management issues,
ed networks (PON) [4], deep space communication , but an efficient buffer is still an important issue.
wildlife and habitat monitoring[5],[6]. The rest of paper is arranged as follows, Section 2
DTN routing protocol [5]build the network infrastruc- present existing buffer managment policies. Section 3 dis-
ture through intermittently connected mobile nodes, cusses router under observation, performance metrics are
where each node store the message in its buffer, carries in Section 4, Section 5 depicts approach and T-DROP al-
the message while it moves and forward it when encoun- gorithm while section 6 is about simulation setup and
ter with other nodes. results with conclusion in section 7.
Because of frequent topology changes, disconnections,
limited contact opportunities, long delays and node mo-
2 EXISTING BUFFER MANAGEMENT POLICIES
bility, routing in DTN environment has become the most
challenging issue. DTN routing schemes can be divided When nodes under DTN resource constrained (buffer)
in to two categories, single copy and multi copy. In single network communicates the congestion arise frequently.
copy scheme [7] only one copy of message is forwarded Hence the issue is which message from congested buffer
by intermittently connected nodes until it find its destina- will be dropped to continue transmission.
tion. Multi copy scheme flood the message copies to dif-
ferent path which increase the robustness and delivery 2.1 Drop Random (DR)
probability. Most of routing strategies based on replica- The selection of message to be dropped is random.
tion of message and therefore multi-copy in nature [8],
[9], [10] and [11]. 2.2 Drop –Least-Recently-Received (DLR)
Despite a huge contribution that have been invested in The message with the long stay time in buffer will be
design of DTN routing protocols, It is evicted that in sto- dropped. The idea is that the packet with in buffer for
rage of message for long time and its replication [12], [13] long time has less probability to be passed to other nodes.
make buffer run out of capacity and effect the perfor-
2.3 Drop oldest (DOA)
———————————————— The message with the shorted remaining life time (TTL)
 Qaisar Ayub is with Universiti Teknologi Malaysia (UTM),Faculty of in network is dropped. The idea of dropping such packet
Computer Science & Information System,Department of Computer System is that if packet TTL is small, it is in the network for long
& Communication, Skudai - Johor, 81310, Malaysia
time and thus has high probability to be already deli-
 Sulma  Rashid  is  with  Universiti  Teknologi  Malaysia  (UTM),Faculty  of 
vered.
Computer Science & Information System,Department of Computer System 
& Communication, Skudai ‐ Johor, 81310, Malaysia  
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 10, OCTOBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 47

2.4 Drop Last (DL)


It drops the newly received message.

2.5 N-Drop
In N-Dropt [9], the message that does N number of for-
warding will be selected to drop.

2.6 Drop largest (DLA)


In Drop Larges (DLA) large size message will be
dropped.[17]

2.7 MOFO - Evict most forwarded first Fig 2 Prophet router


The message that has been forwarded to maximum num‐
ber of times will be dropped first. [8] 
Fig  2  depicts  the  flow  of  prophet  router,  assume  node A 
want  to  transmit  the  message  to  destination  D.  Node  A 
3 PROTOCOL UNDER EVOLUTION will  communicate  with  its  neighbor  C,B  to  check  their 
encounter  probability  with  node  D.  We  can  see  that  pre‐
dictability  P(B,D)>P(C,D)  node  A  will  forward  the  mes‐
3.1 Epidemic router
sage to node B. Hence prophet has higher delivery proba‐
In epidemic routing[8] application messages are flooded
bility  with  much  lower  communication  overhead  then 
to the relay nodes called carriers ,when carrier nodes
epidemic. 
while moving comes in contact with another connected
proton of network , it forward the message to additional
island of nodes. This iteration of forwarding makes sure 4 PERFORMANCE METRICS
the delivery of message to its destination.
4.1 Delivery probability
It is the ratio of message received over message send.
High probability means that more messages are delivered
to the destination.

Fig 1 Epidemic router

Fig 1 depicts the flow of epidemic router, when Node .


A, B comes in the transmission range of each other A ex-
change its summery vector SVA with B. Summery vector 4.2 Overhead
is compact representation about buffered messages and is It is the negation of number of messages relayed to num-
locally maintained for each node. After exchange of ber of message delivered. Low value of overhead means
summery vector Node B perform a logical and operation less processing required delivering the relayed messages.
between the negations of its summery vector to filter out Mathematically it can be represented as
messages not buffered at B. B then transmits a request
summery vector SVR to A which then start the transmis-
sion of requested messages. This process continues as Objective of algorithm is to minimize the value of over-
when node B comes in contact with other node. This re- head.
dundant replication increases the delivery probability on
the cost of buffer space and transmission time.
4.3 Buffer time average
It is Sum of time spend by l message(s) in buffer divided
3.2 Prophet router by message send. Mathematically can represend as
Epidemic [8] replicates messages to all encountered peers, 
while  Prophet  [16]  tries  to  estimate  which  node  has  the 
highest “likelihood” of being able to deliver a message to 
the final destination based on node encounter history. 
Prophet router uses the metric called delivery predictabil‐
ity  which  is  updated  for  each  node  on  encounters.    If 
node  A  wants  to  transmit  the  message  to  destination  D  4.4 Hop count average
then  node  A  look  in  neighbors  nodes  and  forward  the  It is the Mean hops which a message takes to reach its
message to node which has high encounters with D. destination.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 10, OCTOBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 48

5 APPROACH
Check available buffer space
Fig 3 depicts the snapshot of epidemic routing protocol
       for each message M  in MRi 
under congested DOA queue mode where nodes B has
      Bui = getsize (M)+Bui; 
transmited its summery vector SVB to node A. Node A
        end loop; 
then send request summery vector SVR to node B for
            BAi = Bi  ‐  Bui 
transmission of missing packets as eq(1).
      if SMN (source or Relay) < BAi THEN 
    ADD (MRiBAi, MN) 
SVR = M8, M21, M32, M43, M76, M90
  return true; //Message successfully store in buffer 
      end if; 
 // If SMN  > BAi drop some message to make room for      
 While (SMN >BAi) 
  begin 
         //function call to drop threashhold message 
           T_M = SearchThreashHold (); 
           If T_M  is null then 
Fig 3 Messge drop with DOA
               Return false; // no drop 
             end if; 
)
         DropMessage (T_M)) 
Nodes B first transmit the message M8 to node A. In or-
         BAi = sizeof (T_M) +BAi; 
der to accommodate M8, Node A will drop (M1, M2, M3)
         End loop;  
resulting more drop and raise overhead.
          ADD(MRBAi, MN) 
        return true;//message stored in buffer 
Figure 4 depicts the drop of message with T drop (Thre-
 end while; 
shold drop), where the message will be dropped only if
 end; //end of method 
its size fall with in a range of threshold.
 
int SearchThreashhold() 
 begin 
   for each  message M in MR 
       if Size of(M ) between T 

                T_M = M; 
                retrun T_M;  //return message; 
                end if 
Fig 4 Message drop with T-Drop
            end loop; 
  return null; // no threashhold message found 
When Node B transmits the message M8, node A will
           end;//end of method 
check its queue for available buffer space. As we can see
 
that the queue of Node A is congested and it needs to
drop a message. Assume the range of threshold T is be-
tween 400kb and 900kb therefore node A will not drop
any message. This technique give high buffer time to
small messages results in less message drop ,overhead, 6 SIMULATION AND RESULTS
hop count with high delivery probability and buffer time All the experiments were performed using ONE Simula-
average and improve protocols in term of QOS. tor. The ONE [16] Simulator is a discrete event simulator
written in Java. The main objective of simulator is to im-
4.5 Algorithm plement DTN (store-carry-forward) of message for long
TABLE 1 time, where the probability of disconnections and failures
VARIABLE AND MEANINGS is very high.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 10, OCTOBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 49

Fig 5 Drop message with respect to time

In fig 7 we conclude the effect of drop messages with T- Fig 7 Buffer time average with respect to time
Drop and existing DOA under epidemic and prophet
routing protocols. In order to increase the accuracy we
have perform the simulation for different time spans. We In fig 9 we have observe the occupancy of message buffer
can see that proposed T-Drop policy reduce the message time by executing simulation at different time lengths
drop to a huge extent and by increasing simulation time it with existing FIFO and proposed T-drop. We can see with
is getting better. We can examine that FIFO policy cased proposed T-Drop policy messages stay in the buffer for
higher drops while additional loss is consumption of re- long time. This storage of message and prevention against
sources gained by the dropped message during its mobili- drop makes most of messages to be delivered to destina-
ty. T-Drop has control this ambiguity by defining a tion i.e high delivery probability as proved in fig 8.
Threashhold value, if queue is congested only threash-
hold message will be dropped otherwise no drop will
occur.

Fig 8 Hop count average with respect to time


In fig 10 we simulated the result of T-Drop and DOA for
epidemic and prophet router under hop count average.
Fig 6 Delivery probability with respect to time The increase in hop count reflects the fact that message
has consumed more resources to reach its destination.
Fig 8 demonstrates the comparison of delivery probabili- The minimum value of hop count confirms less overhead
ty for DOA and T-Drop queue mode under epidemic and and delivery delay.
prophet router. We can see that at each simulation length We can observe that with T-Drop policy the hop count
proposed queue mode T-Drop deliver more messages as average for both routers has reduced at a constant ratio.
compared to existing DO. Further the raise in delivery In FIFO when ever congestion arise it continue to drops
probability for T-Drop policy is becoming more stable. the messages to free space for new message resulting
Moreover, as simulation time increases both routers with more relay and increase in hop count average. While in T-
T-Drop improve and enhance the performance but in case Drop threashhold value blocks this redundant drop and
of prophet router the delivery probability raise more provides prevention against unnecessary relay.
quickly because it use delivery predictability parameter
for relay nodes.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 10, OCTOBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 50

copy routing in intermittently connected mobile networks,” in


Proc. IEEE Conf. Sensor and Ad Hoc Communications and
Networks (SECON),2004, pp. 235–244.
[8] Vahdat and D. Becker, “Epidemic routing for partially con-
nected ad Hoc networks,” Duke Univ., Durham, NC, Tech. Rep.
CS-200006, Apr.2000.
[9] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine. MaxProp:
Routing for Vehicle-Based Disruption- Tolerant Networks. In
Proc. IEEE Infocom, April 2006.
[10] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine. MaxProp:
Routing for Vehicle-Based Disruption- Tolerant Networks. In
Proc. IEEE Infocom, April 2006. 
[11] A.Lindgren, A. Doria, and O. Schelen, “Probabilistic routing in
intermittently connected networks,” SIGMOBILE Mobile Compu-
Fig 9 Overhead ratio with respect to time ting and Communication Review, vol. 7, no. 3, 2003.
[12] Amir Krifa, Chadi Barakat, Thrasyvoulos Spyropoulos, “Op-
In fig 11 we find the effect of T-Drop and DOA under timal Buffer Management Policies for Delay Tolerant Net-
epidemic and prophet router with respect to overhead works,” IEEE Conference on Sensor, Mesh and Ad Hoc Communica-
ratio. Overhead is consumption of resources (algrothem tions and Networks (SECON 2008), June 2008.
processing) by the relayed messages to reach its destina- [13] Yun Li, Ling Zhao ,Zhanjun Liu,Qilie Liu.” N-Drop Congestion
tion. High ratio of overhead describes more operational Control strategy under Epidemic Routing in DTN.” Research
instructions, network routing information, and retrans- center for wireless information networks,chongqing university
missions of received message. We can see that overhead of posts & Telecommunications ,chongqing 400065,china, pp.
with T-Drop have reduced by avoiding redundant drop 457-460, 2009
and retransmission. While in FIFO policy the overhead is [14] A.indgren and K. S. Phanse, “Evaluation of queuing policies
and forwarding strategies for routing in intermittently con-
increased because of iterative message drop and retrans- nected networks,”in Proc. of IEEE COMSWARE, pp. 1-10, Jan.
mission. 2006.
[15] Sulma Rashid,Qaisar Ayub,”Effective buffer management poli-
cy DLA for DTN routing Protocals under congetion”, Interna-
7 CONCLUSION tional Journal of Computer and Network Security,Vol 2,NO 9,Sep
In this paper we proposed a new buffer management pol- 2010. pp .118-121
icy which drops the message from congested buffer only [16] Homepage of  Oppor tun istic  Network En vir on men t  
if size of existing queued message(s) falls in Threashhold (ONE).  http://www.netlab.tkk.fi/tutkimus/dtn/theone/,  Ac‐
range (T).This technique is called as T-Drop. We prove cessed July 2010. 
through simulations that propose T-Drop policy minim-
 
ize the message in term of drop, hop count average, over-
Authors Profile
head while increase the QOS in terms of delivery proba-
 
bility as compared to existing DOA. Qaisar Ayub
He has obtained his MCS Computer Science degree
References in 2005 from Comsat Institute of Information Technol-
[1] M. Papadopouli and H. Schulzrinne, “Seven degrees of separa- ogy Pakistan. And BCS (Hons.) computer science
tion in mobile ad hoc networks,” in Proc. IEEE GLOBECOM, from Allama Iqbal Open University Pakistan in 2003.
He has 5 years of experience in conducting profes-
2000. sional trainings (Oracle, Java,) and software devel-
[2] Disruption Tolerant Networking. [Online]. Available: opment. As a part of this paper he is working on QOS in DTN
http://www.darpa.mil/ato/solicit/DTN/ routing.
[3] H. Wu, R. Fujimoto, R. Guensler, and M. Hunter, “MDDV:
Mobilitycentric data dissemination algorithm for vehicular Sulma Rashid
networks,” in Proc.ACM SIGCOMM Workshop on Vehicular Ad She has received her MS Degree in computer science
Hoc Networks (VANET),2004  in 2007 from IQRA University Islamabad Pakistan and
MCS degree in 2001 from UAAR Pakistan. She has 10
[4] P. Hui, A. Chaintreau, J. Scott, R. Gass, J. Crowcroft, and C.
years of teaching experience. Her areas of interest are
Diot,“Pocket switched networks and human mobility in confe- DTN, Adhoc, security, Network programming, Operat-
rence environments,”in Proc. ACM SIGCOMM Workshop on De- ing system, wireless networks and MANETS. As a part of this paper
lay Tolerant Networking(WDTN), 2005. she is working on Optimizing and forwarding research issues in DTN
routing.
 
[5] P.  Juang,  H.  Oki,  Y.  Wang,  M.  Martonosi,  L.  S.  Peh,and  D.  Ru‐
benstein.  Energy‐efficient  computing  for  wildlife  tracking:  de‐
sign tradeoffs and early experiences with zebranet. In Proc. AS‐
PLOS’02, Oct. 2002. 
[6] Sensor  networking  with  delay  tolerance 
(sendt).http://down.dsg.cs.tcd.ie/sendt/. 
[7] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Single-

Das könnte Ihnen auch gefallen