International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013
Uncertain Straight Path Thrashing in Disruption
Lenient Networks
K. Ramesh Babu#1, K. Suresh Babu#2
#1
II Year M.Tech Student, #2Associate Professor
#1, #2
CSE Department, Vasireddy Venkatadri Institute of Technology,
Guntur, Andhra Pradesh, India.
Abstract----- Data streaming over Disruption-Lenient Networks
(DLN) is a challenging task considering jointly the specific
characteristics of DLN environments, the demanding nature of
streaming applications and their wide applicability. The Bundle
Streaming Service (BSS) as a framework to improve the
reception and storage of data streams. This framework exploits
the characteristics of Disruption Lenient Networks to allow for
reliable Disruption-Lenient streaming. Here, we present a
simple usage scenario along with the proposed framework and
evaluate it experimentally at a preliminary stage which, however,
suffices to demonstrate its potential suitability for both
terrestrial and Space environments. Based on the observations
about human mobility traces and the findings of previous work,
we introduce a new metric called Uncertain intermeeting time,
which computes the standard intermeeting time between two
nodes relative to a meeting with a third node using only the local
knowledge of the past contacts. We propose Uncertain Straight
Path Thrashing (USPT) protocol that routes the messages over
Uncertain Diminished paths in which the cost of links between
nodes is defined by uncertain intermeeting times.
Keywords— USPT, Uncertain Intermeeting time, Straight
Intermeeting time, DLN.
.
I.
INTRODUCTION
After several years of systematic research in various
aspects of Disruption Lenient Networking (DLN) such as
routing, transport protocols and convergence layers, DLN
technology has reached a higher level of maturity. The
development of a reliable set of working solutions and
associated standards under the auspices of the Consultative
Committee for Space Data Systems (CCSDS) and the Internet
Research Task Force's (IRTF's) DLN research group [1] has
boosted the applicability of DLN architectures, which now
present themselves as prominent solutions for global internetworking. Based on that progress, several studies [2], [3], [4]
promote the benefits of DLN architectures [5] and highly
suggest their use in disruptive environments through the
Bundle protocol [6], which encodes most functionality that an
overlay network re- quires. That is, there is an assumption of
'global reach ability’ in the Internet, Many of the applications
that users have come to appreciate also rely on the round trip
time (RTT) for data packets being quite small, so that, for
example, a screen display for a web page can be built up from
information retrieved by several separate requests to one or
more servers or information stores. Thrashing in Disruption
ISSN: 2231-5381
Lenient Networks (DLN) is a challenging problem because at
any given time instance, the probability that there is an end-toend path from a source to a destination is low.
Mobile Ad-hoc Networks (MANETs) are closely related
to DLNs since they share several common characteristics such
as network disruptions, high error rates and variable capacity
links. A substantial amount of prior works that address
several data streaming issues have already been proposed for
MANETs. In general, the majority of the efforts are moving
in two main directions; i) efficiency improvement and ii)
redundancy. Among the most popular approaches suggested
so far for improving efficiency are: i) the dynamic
optimization of data coding, throughout the streaming session,
so that the encoding bitrates does not surpass the available
bandwidth of the network [8], ii) routing through multiple
paths in order to increase delivery probability [9], iii) packet
prioritization to minimize queuing delay and iv) specially
adapted transport layer mechanisms that aim in reducing
recovery delay of lost data. Redundancy on the other hand, is
achieved through the use of FEC codes or by applying content
summarization and error spreading techniques in order to
provide error resilience.
Thrashing in DLN’s utilize a paradigm called storecarry-and-forward. When a node gets a message from one of
its contacts, it stores the message in its buffer and carries the
message until it encounters another node which is at least as
useful (in terms of the release) as itself. Then the message is
forwarded to it. Recent studies on routing problem in DLN’s
have focused on the analysis of real mobility traces (human
[11], vehicular [12] etc.). First, rather than being memory less,
the pair wise intermeeting times between the nodes usually
follow a log-normal distribution [13] [14]. Therefore, future
contacts of nodes become dependent on the previous contacts.
Second, the mobility of many real objects is non-deterministic
but cyclic [15]. Hence, in a cyclic MobiSpace [15], if two
nodes were often in contact at a particular time in previous
cycles, then they will most likely be in contact at around the
same time in the next cycle.
To illustrate the benefits of the planned metric, we
assume it for the Straight path based routing algorithms [7],
[10] intended for DLN’s. We propose Uncertain Straight Path
Thrashing (USPT) protocol in which standard uncertain
intermeeting times are used as link costs rather than typical
intermeeting times and the messages are routed over
http://www.ijettjournal.org
Page 3894
International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013
Uncertain Straight Paths (USP). We evaluate USPT protocol
with the accessible Straight Path (SP) based thrashing protocol
through real trace- driven simulations. The results demonstrate
that USPT achieves higher delivery rate and lower end-to-end
delay compared to the shortest path based routing protocols.
This shows how well the straight intermeeting time represents
inter- node link costs (in the context of Thrashing) and helps
making effective forwarding decisions while Thrashing a
message.
II. UNCERTAIN INTERMEETING TIME
An analysis of real mobility traces has been done in
different environments (office [13], conference [16], city [19],
skating tour [14]) with different objects (human [11], bus [12],
zebra [20]) and with variable number of attendants and led to
significant results about the aggregate and pair wise mobility
characteristics of real objects. Recent analysis [13], [14], [16]
on real mobility traces have demonstrated that models
assuming the exponential distribution of intermeeting times
between pairs of nodes do not match real data well. Instead up
to 99% of intermeeting times in many datasets is log-normal
distribution.Further properly, if C is the random variable
representing the intermeeting time between two nodes, P (C >
a + b | C > b)  P (C > a) for a, b > 0.
To get improvement of such information, we suggest a
new metric called Uncertain intermeeting time that calculates
the intermeeting time among two nodes virtual to a
congregation with a third node using only the limited
information of the past contacts. Consider the sample cyclic
MobiSpace with three objects illustrated in Figure 1. The
frequent movement rotation is 11 time units, so the separate
probabilistic contacts between A and B happens in every 13
time units (1, 14, 27, 40 ...) and between B and C in every 8
time units (2, 10, 18, 26 ...). The standard intermeeting time
between nodes B and C indicates that node B can forward its
message to node C in 9 time units. However, the uncertain
intermeeting time of B with C relative to prior meeting of
node A indicates that the message can be forwarded to node C
within one time unit.
 τA(B): Average time that elapses between two
consecutive meetings of nodes A and B. Obviously when
the node connections are bidirectional, τA(B) = τB(A).
 τA(B|C): Average time it takes for node A to meet node B
after it meets node C. Note that, τA (B|C) and τB(A|C) are
not necessarily equal.
 A: M × M matrix where A(a, b) shows the sum of all
samples of Uncertain intermeeting times with node b
relative to the meeting with node a. Here, M is the
neighbour count of current node (i.e. M (x) for node x).
 C: M × M matrix where C(a, b) shows the total number of
Uncertain intermeeting time samples with node b relative
to its meeting with node a.
 βa: Total meeting count with node a.
In Algorithm 1, each node first add up times expired
between repeating meetings of one neighbour and the
meeting of another neighbour. Then it divides this total by
the number of times it has met the first neighbour prior to
meeting the second one. While computing standard and
uncertain intermeeting times, we ignore the edge effects [12]
by including intermeeting times of atypical meetings. That
means that we include the values of τA(B) for the first and
last meetings of node B with node A. Likewise, we include
the values of τA(B|C) for the first meeting of node A with
node C and the last meeting of that node with node B.
Fig 2: Example convention periods of node A with nodes B and C. While the
values in the upper part are used in calculation of τA(B|C) and the
values in lower part are used in the calculation of τA(C|B).
Thrashing decisions can be made at three different points
in an SP based Thrashing: i) at source, ii) at each hop, and iii)
at each contact. In the first one (source Thrashing), SP of the
message is decided at the source node and the message
follows that path. In the second one (per-hop Thrashing),
when a message arrives at an intermediate node, the node
determines the next hop for the message towards the
destination and the message waits for that node. Finally, in the
third one (per-contact Thrashing), the Thrashing table is
recomputed at each contact with other nodes and the
forwarding decision is made accordingly. In these algorithms,
Fig 1: An objective recurring MobiSpace with an ordinary
utilization of recent information increases from the first to the
In a DLN, each node can calculate the normal of its last one so that better forwarding decisions are made; however,
more processing resources are used as the Thrashing decision
standard and restricted intermeeting times with other nodes
using its contact history. In Algorithm 1, we show how a node, is computed more frequently.
Consider the sample meeting times of a node A with its
x, can calculate these metrics from its previous meetings. The
neighbors
B and C in Fig 2. Node A meets with node B at
notations we use in this algorithm are listed below with their
times
{6,
20,
32, 38} and with node C at times {13, 17, 28, 42}.
meanings:
Following the procedure described above, we find that τA
(B|C) = (7 + 8 + 4 + 10)/4 = 7.25 time units and τA (C|B). = (7
+ 3+ 4)/3 = 4.66 time units.
ISSN: 2231-5381
http://www.ijettjournal.org
Page 3895
International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013
nodes. The neighbors of a node m are denoted with N(m) and
the edge sets are given as follows:
Algorithm 1: update (node s, time t)
1: if s is seen first time then
2: firstTimeAt[s] ← t
3: else
4: increment βs by 1
5: lastTimeAt[s] ← t
6: end if
7: for each neighbor b  M and b ≠ s do
8: start a timer tsb
9: end for
10: for each neighbor b  M and b ≠ s do
11: for each timer tbs running do
12:
X[b][s] += time on tbs
13:
increment Y[b][s] by 1
14: end for
15: delete all timers tbs
16: end for
17: for each neighbor a  M do
18: for each neighbor b  M and b ≠ a do
19:
if X[b][a] ≠0 then
20:
τX(a|b) ← X[b][a] / Z[b][a]
21:
end if
22: end for
23: τX(a) ← (lastTimeAt[a] − firstTimeAt[a]) / βa
24: end for
E= Eu
 Eb
Eb = {(a, b) |  b  N(m)} where w(a, b) = τa (b) = τb (a)
Eu = {(a, b) |
 b,c N(m) and b≠c} where w(a,b) =τa(b|c)
In Figure 3, we illustrate a sample DLN graph with four
nodes and nine edges. Of these nine edges, three are
bidirectional with weights of standard intermeeting times
between nodes, and six are unidirectional edges with weights
of uncertain intermeeting times.
Fig 3: The graph of a sample DLN with four nodes and nine edges in total.
III. UNCERTAIN STRAIGHT PATH THRASHING
a) Overview:
Straight path thrashing protocols for DLN’s are based on
the designs of routing protocols for traditional networks.
Messages are forwarded through the Straight paths between
source and destination pairs according to the costs assigned to
links between nodes. Furthermore, the dynamic nature of
DTN’s is also considered in these designs. Two common
metrics used to define the link costs are minimum expected
delay (MED [7]) and minimum estimated expected delay
(MEED [10]). They compute the expected waiting time plus
the transmission delay between each pair of nodes. However,
while the former uses the future contact schedule, the latter
uses only observed contact history.
b) Network Model:
The receiver’s application is built using the BSS library,
which initiates a back- ground thread that receives all the
bundles. Whenever that thread receives a bundle, it inserts the
bundle into the BSS database (in creation-time order, for
replay on demand) and it also checks bundle’s creation time in
order to decide, based on the above-described rule, if it will
pass the bundle to an application-provided call-back function
for real-time display or to other stream processing. We model
a DLN as a graph G = (V, E) where the vertices (V) are mobile
nodes and the edges (E) represent the connections between
these nodes. However, different from previous DLN network
models [7], [10], we assume that there may be multiple
unidirectional (Eu) and bidirectional (Eb) edges between the
ISSN: 2231-5381
c) Uncertain Straight Path Thrashing:
Our algorithm basically finds Uncertain Diminished
Paths (USP) for each source-destination pair and routes the
messages over these paths. We define the USP from a node m0
to a node pz as follows:
USP (p0, pz) = {p0, p1, p2… pz-1, pz
|
Rp 0( p
1 t)
+
z 1
 n ( p
k
pk  1) is minimized}
k 1
k 1
Here, t represents the time that has passed since the last
meeting of node p0 with p1 and R p 0 ( p 
1 t ) is the expected
residual time for node p0 to meet with node p1 given that they
have not met in the last t time units. R p 0 ( p 
1 t ) can be
computed with parameters of distribution representing the
intermeeting time between p0 and p1. Assume that node g
observed n intermeeting times with node h in its past. Let
 1g ( h ),  2g ( h ),... ng ( h ) denote these values. Then:
n
Rg(ht)

=
f gi ( h )
i 1
where
i
   g ( h )  t 
  h   t  if  gi  h  t
f gi ( h )  
 0  O th e r w is e
 ig
Each node forms the aforementioned network model and
collects the standard and uncertain intermeeting times of other
nodes between each other through epidemic link state protocol
http://www.ijettjournal.org
Page 3896
International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013
as in [10]. However, once the weights are known, it is not as
easy to find USP’s as it is to find SP’s. Consider Figure 5
where the USP(A, E) follows path 2 and USP(A, D) follows
(A, B, D). This situation is likely to happen in a DTN, if
τD(E|B) ≥ τD(E|C) is satisfied. Running Dijkstra’s or
Bellman-ford algorithm on the current graph structure cannot
detect such cases and concludes that USP(A, E) is ( A, B, D,
E).
Given a DLN graph G = (V, E), we obtain a new graph
G’ = (V’, E’) where:
V’  V  V and E’  V’  V’ where
V '{( a b )
  b  N (m )}a nd  E '{( a b , c u )
 m  u }


 w h ere  w ' ( a b , c u   


 c 
 b )if b  i
 c )o th erw ise
IV. REPLICATIONS
In this section, we describe the details of our simulations
through which we compare the proposed Uncertain Straight
Path Thrashing (USPT) algorithm with standard Straight Path
Thrashing (SPR). For a simulation run, we generated 5000
messages from a random source node to a random destination
node at each t seconds. In Roller Net, since the duration of
experiment is short, we set t =1s, but for Cambridge data set,
we set t =1min.
We assume that the nodes have enough buffer space to
store every message they receive, the bandwidth is high and
the contact durations of nodes are long enough to allow the
exchange of all messages between nodes. These assumptions
are reasonable in today’s technology and are also used
commonly in previous studies [18]. Moreover, we compare all
algorithms in the same conditions, and a change in the current
assumptions is expected to affect the performance of them in
the same manner. We ran each simulation 10 times with
different seeds but the same set of messages and collected
statistics after each run. The results plotted in Figures 7 and 8
show the average of results obtained in all runs.
The graph can be seen also as the representation of the
capacity-delay region achievable in the two cases. Note that
this region shows some performance limitations of the DLN
considered in the experiment; this is coherent with previous
work [15] and due to the fact that time schedule for public
transportation is inherently designed to reduce contacts among
the buses. Figure 5 shows the delivery rates achieved in USPT
and DPR algorithms with respect to time (i.e. TTL of
messages) in RollerNet traces [18]. Clearly, USPT algorithm
delivers more messages than DPR algorithm. Moreover, it
achieves lower average delivery Disruption than SPR
algorithm.
Note that the edges in Eb (in G) are made directional in G’ and
the edges in Eu between the same pair of nodes are separated
in E’. For example, for a path A,B,C,D in G, an edge like
(CD,DA) in G’ cannot be chosen because of the edge settings
in the graph. Hence, only the correct τ values will be added to
the path calculation. To solve the USP problem however, we
add one vertex for source S (apart from its permutations) and
one vertex for destination node D. We also add outgoing
edges from S to each vertex (iS)  V’ with weight RS(i|t).
Furthermore, for the destination node, D, we add only
incoming edges from each vertex ij  V’ with weight τi(D|j).
In Figure 4, we show a sample transformation of a clique
of four nodes to the new graph structure. In the initial graph,
all mobile nodes A to D meet with each other, and we set the
source node to A and destination node to D (we did not show
the directional edges in the original graph for brevity). The
focus of this paper is an improvement of the current design of
the Straight Path (SP) based DLN Thrashing algorithms.
Therefore we leave the elaborate discussion of some other
issues in SP based Thrashing (complexity, scalability and
Thrashing type selection) to the original studies [7] [10].
We believe that in current DLN’s, wireless devices have
enough storage and processing power not to be unduly taxed
with such an increase. Moreover, to lessen the burden of
collecting and storing link weights, an asynchronous and
distributed version of the Bellman-Ford algorithm can be used,
Fig 5: Message delivery ratio vs. time in RollerNet traces.
as described in [17]. In G’, |V ‘| = O (|V|2) and |E’| = O (|V3|)
These results show that the uncertain intermeeting time
3/2
3
= |E| . Therefore Dijkstra’s algorithm will run in O (|V |)
represents link cost better than the standard intermeeting time.
(with Fibonacci heaps) while computing standard Straight
BSS manages to reduce the total requested time of receiving
paths (where edge costs are standard intermeeting times) takes
5000 frames by almost 80% in the worst case. In Space
2
O (|V| ).
environments, where LTP “red” transmission is used in place
of TCP, BSS achieves better results only in cases where the
error rate of the channel is above 10%. Furthermore, based on
a different set of experiments that due to lack of space we
cannot present here, we note another interesting property of
BSS: it manages to reduce the total number of out-of-order
received packets in comparison with the normal ION
configuration using LTP alone.
Therefore, in USPT, more effective paths with similar
average hop counts are selected to route messages.
Consequently, higher delivery rates with lower end-to-end
Disruptions are achieved. In SPR and USPT algorithms here,
Fig 4: Graph Transformation to solve USP with 4 Nodes where A is the
we used source-Thrashing and let the messages follow the
source and D the destination node.
paths which are decided at the source nodes [20].
ISSN: 2231-5381
http://www.ijettjournal.org
Page 3897
International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013
V. CONCLUSION
In this paper, we introduced a new metric called Uncertain
intermeeting time inspired by the results of the recent studies
showing that nodes’ intermeeting times are not memory less
and that motion patterns of mobile nodes are frequently
repetitive. Then, we looked at the effects of this metric on
Straight path based Thrashing in DLN’s. For this purpose, we
updated the Straight path based Thrashing algorithms using
uncertain intermeeting times and proposed to route the
messages over Uncertain Straight paths.
Finally, we ran simulations to evaluate the proposed
algorithm and demonstrated the superiority of USPT protocol
over the existing Straight path thrashing algorithms. For this,
we plan to use probabilistic context free grammars (PCFG)
and utilize the construction algorithm presented in [26]. Such
a model will be able to hold history information concisely, and
provide further generalizations for unseen data.
VI. REFERENCES
Delay
tolerant
networking
research
group,
http://www.dtnrg.org.
[2] T. Spyropoulos, K. Psounis,C. S. Raghavendra, Efficient
routing in inter- mittently connected mobile networks: The
single-copy case, IEEE/ACM Transactions on Networking,
vol. 16, no. 1, Feb. 2008.
[3] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine,
MaxProp: Routing for Vehicle-Based Disruption- Tolerant
Networks, In Proc. IEEE Infocom, April 2006.
[4] A. Vahdat and D. Becker, Epidemic routing for partially
connected ad hoc networks, Duke University, Tech. Rep.
CS-200006, 2000.
[5] T. Spyropoulos, K. Psounis,C. S. Raghavendra, Efficient
routing in inter- mittently connected mobile networks: The
multi-copy case, IEEE/ACM Transactions on Networking,
2008.
[6] Y. Wang, S. Jain, M. Martonosi, and K. Fall, Erasure
coding based routing for opportunistic networks, in
Proceedings of ACM SIGCOMM workshop on Delay
Tolerant Networking (WDTN), 2005.
[7] S. Jain, K. Fall, and R. Patra, Routing in a delay tolerant
network, in Proceedings of ACM SIGCOMM, Aug. 2004.
[8] T. Spyropoulos, K. Psounis,C. S. Raghavendra, Spray and
Wait: An Efficient Routing Scheme for Intermittently
Connected Mobile Networks, ACM SIGCOMM Workshop,
2005.
[9] A. Lindgren, A. Doria, and O. Schelen, Probabilistic
routing in in- termittently connected networks,
SIGMOBILE Mobile Computing and Communication
Review, vol. 7, no. 3, 2003.
[10] E. P. C. Jones, L. Li, and P. A. S. Ward, Practical routing
in delay tolerant networks, in Proceedings of ACM
SIGCOMM workshop on Delay Tolerant Networking
(WDTN), 2005.
[11].Internet Research Task Force Delay Tolerant Networking
Research
Group,
http://www.dtnrg.org/,
http://www.irtf.org/charters/dtnrg.html
[12]. Farrell, S., Cahill, V., Geraghty, D., Humphreys, I.,
McDonald, P.: When TCP Breaks: De- lay and Disruption
Tolerant Networking. IEEE Internet Computing, 72–78
(2006)
[1]
ISSN: 2231-5381
[13] T. Spyropoulos, K. Psounis, and C. Raghavendra, Spray
and Focus: Efficient Mobility-Assisted Routing for
Heterogeneous and Correlated Mobility, In Proceedings of
IEEE PerCom, 2007.
[14] E. Daly and M. Haahr, Social network analysis for
routing in disconnected Disruption-Lenient manets, In
Proceedings of ACM MobiHoc, 2007.
[15] P. Hui, J. Crowcroft, and E. Yoneki, BUBBLE Rap:
Social Based Forwarding in Disruption Lenient Networks,
In Proc. of ACM MobiHoc, 2008.
[16] A European Union funded project in Situated and
Autonomic Communications, www.haggleproject.org.
[17] D. Bertsekas, and R. Gallager, Data networks (2nd ed.),
1992.
[18] C. Liu and J. Wu, An Optimal Probabilistically
Forwarding Protocol in Disruption Lenient Networks, in
Proceedings of MobiHoc, 2009.
[19] J. Leguay, A. Lindgren, J. Scott, T. Friedman, J.
Crowcroft and P. Hui, CRAWDAD data set upmc/content
(v.
2006-11-17),
downloaded
from
http://crawdad.cs.dartmouth.edu/upmc/content, 2006.
[20] Y. Wang, P. Zhang, T. Liu, C. Sadler and M. Martonosi,
http://crawdad.cs.dartmouth.edu/princeton/zebranet,
CRAWDAD data set princeton/zebranet (v. 2007-02-14),
2007.
http://www.ijettjournal.org
Page 3898