ACHIEVING LOW END-TO-END LATENCY WITH A
HANDOFF-BASED DETERMINISTIC ROUTING PROTOCOL
(HDRP) IN DELAY-TOLERANT NETWORKING
Anika Aziz and Shigeki Yamada
National Institute of Informatics, The Graduate University for Advanced Studies
2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo 101-8430, Japan
Keywords: DTN, Bundle, Handoff, HDRP, Latency.
Abstract: DTN is able to adapt any mobility environment where any mobile routers and terminals are combined
owing to the DTN’s flexible hop-by-hop routing schemes. The existing DTN protocols like Epidemic,
Prophet and Spray-and-Wait protocols rely on the message distribution mechanism where each DTN node
produces one or more message copies. They can naturally adapt to the mobile situation where the
destination node moves from an old connection endpoint to a new connection endpoint because any
message copy may luckily be able to reach the new connection endpoint where the mobile node is newly
connected. These protocols suffer from long latencies because message copies are not immediately
forwarded until any suitable condition for forwarding is met. To solve these problems, we propose a
Handoff–based Deterministic Routing Protocol (HDRP) that makes the best use of general handoff
mechanisms intended for the IP network. This handoff mechanism includes the registration of locations by
mobile nodes and backward propagation and caching of these locations over the experienced route. Our
simulation results indicate that HDRP outperforms other existing protocols especially in terms of end-to-end
latency.
1 INTRODUCTION
The architecture and protocols devised to be used in
Delay and Disruption Tolerant Networking (DTN)
are well suited for mobile and extreme environments
lacking continuous connectivity (Ott, 2005). The
DTN architecture is featured by dynamic hop-by-
hop routing and the custody transfer mechanism.
The Custody Transfer concept refers to the
acknowledged delivery of a message from one DTN
hop to the next and the corresponding passing of
reliable delivery of the responsibility. A DTN node
which has taken custody of a Bundle will buffer it
until a suitable next hop is found (Fall, 2003). This
also means that routing should dynamically be
changed, depending on the ongoing situations. So,
the re-routing function is equipped with the DTN
architecture as hop-by-hop routing.
The existing DTN protocols like Epidemic,
Prophet and Spray-and-Wait protocols (Vahdat and
Becker, 2000), (Lindgren and Schelen, 2003),
(Spyropoulos and Raghavendra, 2005) handle
mobile situations, based on the message distribution
mechanism where each DTN node produces one or
more message copies, wait for a suitable condition
for forwarding like meeting a node with higher
delivery predictability in Prophet, then distributes
them possibly on a hop-by-hop basis toward the
destination node, and makes any lucky one of the
copies reach the destination node. Therefore, the
existing DTN protocols can naturally adapt to the
mobile situation where the destination node moves
from an old connection endpoint to a new
connection endpoint because any message copy may
luckily be able to reach the new connection endpoint
where the mobile node is newly connected. However
the existing DTN Protocols suffer from low delivery
ratio because their routing decisions are based on the
local knowledge given by the next hop node. The
existing DTN protocols also suffer from long
latencies because message copies are not
immediately forwarded until any suitable condition
for forwarding is met. These problems suggest that
any global knowledge on end-to-end path from the
source node to the destination node should be more
13
Aziz A. and Yamada S. (2010).
ACHIEVING LOW END-TO-END LATENCY WITH A HANDOFF-BASED DETERMINISTIC ROUTING PROTOCOL (HDRP) IN DELAY-TOLERANT
NETWORKING.
In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 13-22
DOI: 10.5220/0002987800130022
Copyright
c
SciTePress
exploited to enable immediate forwarding that
eliminates any opportunistic waiting.
We assumed a practical environment with fixed
routers and mobile or fixed terminals. Because
when we project the future use of network, we find
that more and more computer and network resources
will be ubiquitously located in fixed locations just
like the Wireless LAN environment. We propose a
Handoff–based Deterministic Routing Protocol
(HDRP) that makes the best use of general handoff
mechanisms intended for the IP network. Here, the
handoff refers to the dynamic event in which
communications are maintained with the network by
transferring the connection to a neighbouring
network access point. The proposed handoff-based
mechanism includes the registration of locations by
mobile nodes and backward propagation and
caching of these locations. When a mobile node
moves to a new location, it registers its location (the
name of the router it belongs to) with the DTN
router and this location information is propagated to
and cached in every DTN router over the
experienced route to update the Proxy List (PL) at
each router. As every DTN router maintains the
connectivity information with adjacent DTN routers
in the PL, the DTN router can select the best
possible next hop for bundles destined to the mobile
node in a deterministic way. Our simulation results
indicate that HDRP outperforms other existing
protocols like Epidemic, PROPHET and Spray-and-
Wait especially in terms of end-to-end latency.
2 MOTIVATION
The mobility environment supported by existing
protocols in DTN is by having forwarding decision
on the basis of local knowledge given by the next
hop node and is done opportunistically depending
on some condition to be fulfilled. These lead to
degraded performances in terms of delivery ratio and
end-to-end latency. The forwarding process can be
improved by making use of the location information
of the mobile node and back propagating and
caching this information on the experienced route of
the mobile node. As every DTN router maintains the
connectivity information with adjacent DTN routers
in the PL, the DTN router can select the best
possible next hop for Bundles destined to the mobile
node in a deterministic way. The novelty of the work
lies in gathering the routing information through the
Hand off process which provides the location
information of the MH to the previous router as part
of the process and hence helps to route the data more
deterministically and quickly to the MH. Moreover
the Back Propagation and caching of this location
information at each of the routers in the experienced
route of the MH provides option to choose the best
route to reach the destination. This feature adds
more dynamic and improved performance to the
HDRP.
3 THE PROPOSED SCHEME
3.1 The System Model
Our system model assumes an environment of
interconnected fixed DTN routers. We have
considered the same network environment as the
WLAN environment which is very practical if we
project the future use of network.
Figure 1: Handoff Bundles from one node to another after
MH changes location.
There are two types of End nodes, Fixed Host
(FH) and Mobile Host (MH) which are
communicating with each other through these
routers. The FHs are connected with DTN routers
and the MHs are moving around following a map
based movement model (Keränen and Kärkkäinen,
March 2009). An MH can communicate directly
with another MH if they come within each others’
range.
If an MH move to a new location it registers its
location to the new router and this location
information is propagated back to the old router. The
old router then hand off the data destined for the MH
to the new router. Our main focus was to propose a
routing protocol which is based on this handoff
technique and at the same time to utilize the hop-by-
hop routing and custody transfer mechanism of the
DTN routers.
The DTN nodes communicate using the Bundle
Protocol for DTN and the protocol data unit is
known as Bundle which is the aggregated messages.
Basically two types of Bundles are used -the data
type and the Status Report (SR) type, latter of which
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
14
is an administrative type of record for sending
acknowledgement to a custody transfer request etc
(Scott and Burleigh, November, 2007).
3.2 The Routing Protocol
Our proposed routing protocol named Handoff-
based Deterministic Routing Protocol (HDRP) has
some key conceptual features and technical features.
This section explains each of these features in detail
along with an example of a particular protocol
sequence.
Every router has a Proxy List (PL) where it
keeps record of all other routers in the network and
the next hop to choose to reach each of the routers.
When an MH moves to a new location it registers its
location with the new router after receiving a beacon
from that router. The registration message, REG
contains [MH, Previous Master (PM)] addresses that
is, the address of the MH and the old router. After
registration is done the new router becomes the
Current Master (CM) of this MH and forwards a
Handoff Message containing (MH, CM) to the
Previous Master.
In the mean time, when the old router senses
(senses periodically whether the MH is within its
range) that the MH has moved away it starts
buffering the bundle destined for that MH. Upon
contacted by the new router (through the handoff
message), the old router handoff the buffered
bundles to the new router and finally the bundles are
forwarded to the MH. The old router also updates
the PL with the (MH, CM) information so that it can
forward the subsequent bundles destined for that
particular MH using this route update. The process
of handing over the buffered data from the old router
to the new router is done through the hop-by hop
routing method of DTN and the bundles are kept in
the buffer of the old router until a new route is
established to the destination with the help of the
custody transfer mechanism.
Depending on the situation whether the direction
of communication is from FH to MH or from MH to
FH and whether the Bundle/SR has been lost during
sending/ receiving process, there can be different
protocol sequences for the handoff.
Figure 2 shows one of those protocol sequences
between different DTN nodes in our model where
MH fails to receive a Bundle due to its movement.
In this diagram, a FH is sending Bundles to a
MH through Router1 (old router). The MH changes
its location and moved to Router2 (new router). The
transfer of the data and SR bundles with specific
functions are shown in the figure. To accomplish the
Figure 2: Protocol sequences showing a handoff situation
and route update by Back propagation.
handoff process successfully, we extend the Bundle
specified in the IRTF standardization in order to
include the auxiliary addresses and its associated
fields in the SR. The Handoff latency has been so far
defined in different ways in different works that deal
with the handoff problem (The ONE), (Manzoni et
al.), (Yavatkar and Bhagawat, 1994). We have
defined the Handoff latency as the time from “when
the MH receives the beacon from the new router” to
“the time when the MH receives the first Bundle
through the newly established route”. Here we have
assumed the best possible case for the arrival of the
beacon that is we have assumed that as soon as the
MH entered into the range of a router it receives a
beacon from that router.
The above figure shows the simplest case when
the old and new routers are only one hop away from
each other. Our protocol also works efficiently
even
if the MH has moved far away that is even if there
are one or more routers in between the old and new
routers. In our system model, we have assumed that
epidemic protocol is used for initial Bundle delivery
to the router that will initiate the handoff. This is the
situation before any MH in the network starts doing
registration or any handoff has taken place. A very
large sized buffer at the routers is assumed to cope
with the situation of buffer overflow for the custody
transfer.
As every DTN router in the HDRP maintains the
connectivity information (through PL) with adjacent
DTN routers, the DTN router can select the best
possible next hop for Bundles destined to the MH in
a deterministic way. The Forwarding mechanism
also plays a vital role in achieving the deterministic
routing: A router always look for a direct connection
while forwarding a bundle. If it is not found then the
router consults the PL and lastly it goes for the
ACHIEVING LOW END-TO-END LATENCY WITH A HANDOFF-BASED DETERMINISTIC ROUTING
PROTOCOL (HDRP) IN DELAY-TOLERANT NETWORKING
15
flooding technique. A Flood List (FL) is maintained
for each of the Bundles to be flooded through all the
connections attached to a HDRP router. Bundles are
sent according to the following priority: (1) Data
from Buffer, (2) Data from Flood List (FL), (3)
Handoff message, and (4) SR (Status Report). As
shown, the data type always gets the priority over
other types of Bundle.
3.3 The Bundle Protocol Extension
The features that we need to implement in our
routing protocol required certain fields in the
standard Bundle block format which is not specified
in the present Bundle Protocol Specification (Scott
and Burleigh, 2007). So, we propose some extension
to the present Bundle Block format. To accomplish
the Handoff process, the MH needs to do registration
with the newly found router after receiving beacon
from it:
During registration the MH will inform the
Current Master (CM) about its Previous Master
(PM) and for this a field should carry the EID
of the Previous Master.
If When the CM will forward a Handoff
message to the PM it needs to inform about the
EID of the new MH that has just completed
registration with it and used to reside with the
old one.
We assume that this type of EID information can be
carried as an auxiliary EID field in the payload
block.
Figure 3: Extension to the Bundle block to support
handoff mechanism in DTN.
As shown in Figure 3, A Bundle normally have
the Primary Block and Payload Block, the extension
Block is optional. The extension to the Bundle block
is carried out in following steps:
The ‘Bundle payload variable’ field normally
carries Data but the field can also carry
Administrative records like special Status
Report.
In this case, a flag in the Primary block
indicates if it is carried in the Bundle payload
field (Scott and Burleigh, 2007).
The ‘Record type code’, ‘Administrative record
flag’ and ‘record content’ field will indicate
that the following information is carried to
accomplish a handoff process.
Finally the status flag and the reason code of
the ‘Record content’ specifies about the
auxiliary address and the auxiliary EID type is
given with the value and meaning at the end of
the Status Report.
The extension in the Bundle block format that we
have proposed will not have any influence on the
operation of the current Bundle protocol.
In Back propagation, the route update
information of a particular MH is propagated back in
its experienced route. The concept is implemented in
the routing protocol. We do not need to make any
change in the Bundle block format to support this
mechanism. The route update propagation and
caching at each of the routers in the experienced
route involve the Handoff message and REG which
is supported by the Bundle extension that we have
proposed earlier.
Our protocol is also suitable to work in an
environment where mobile routers are used. After
receiving the Handoff Message, the Previous Router
(PM) of the MH sends Bundles destined for that MH
to its Current master (CM). If the CM changes its
position by this time due to mobility, the Bundles
can still reach the destination because if the PM fails
to reach the CM by direct connection or PL it will
ultimately go for the flooding technique as the last
resort to reach the CM of the MH. As a result, even
if the CM is a mobile router it will eventually get the
Bundles to be delivered to the MH.
On the other hand, if it happens that the PM
changes its position while the MH is going through
the registration process to its CM, the scenario will
be handled in a similar way as the case when the PM
is multi hop away from the CM.
4 PERFORMANCE EVALUATION
AND ANALYSIS
4.1 Simulation Network Model
Our network model consists of Fixed routers
connected in a certain topology, Fixed end nodes
connected to some of these routers and different type
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
16
of mobile nodes moving around this fixed
infrastructure network to communicate with other
Table 1: Parameters used in the simulation.
Parameter Description (Values)
Node type
Fixed Routers: scenario 1 for 26
Routers; scenario2 for 65 Routers.
End nodes: The numbers were varied
from 124 to 1000.
Connections &
movement
model
Most of the Routers are connected
with each other;
Map based movement model (Keränen
and Kärkkäinen, March 2009)
Mobile node
types
Pedestrians, cars, trams
Transmitting
speed of the
Routers & End
nodes
20M,250k Bytes per second
Transmitting
range of the
Routers & End
nodes
300m
Message
interval&
Message size
Message interval of (1,2000), (25,35),
(10,20) and (1,5) ( per second) with
random distribution;
Message size is randomly varied
between 500KB and 1MB
Buffer size Varied from 10Mbytes to 300Mbytes
Mobility
speed(M)
(Minimum speed, maximum speed)
in m/s of each type of mobile node
groups × no. of hosts in that
group/total no. of nodes; M1,M2,M3
and M4 are calculated by varying the
range of the speed for each mobile
node types and finally by varying all
together.
Simulation time 43K ~= 12 hours
Message TTL
40mins (for discarding messages)
Alive time/Back
Propagated
(BP) Routing
update expired
time/cache time
We varied the value for 5 sec, 30 sec,
160 sec, 600 sec and 3000 sec.
end nodes. Communication is possible between any
pair of end nodes through one or more routers. The
mobile nodes can communicate directly with each
other if they are within the communication range.
Different parameters used in the simulation are listed
in Table 1. We have used the ONE simulator
(Keränen and Kärkkäinen, 2009), (The ONE) to
build and simulate our network model.
We made some modification and do some
extension in the ONE simulator in order to
implement our protocol. Our HDRP Routing
protocol extends the Active Router module used in
the ONE simulator. Fields and methods have been
created to implement the Handoff mechanism which
was not included in any DTN routing algorithm in
ONE before. Handoff Reports have been generated
by extending the Report module.
4.2 Simulation Results
4.2.1 Performance at Different Traffic
Intensity
In ONE simulator we can set different Message
interval values within a minimum and maximum
range. We varied this interval for (1, 2000), (25, 35),
(10, 20) and (1, 5) per second. For each of these
ranges, messages are generated randomly within the
limits which actually indicate how much densely the
messages are generated within our specified
simulation time. That is why here we refer the
different Message interval as the different traffic
intensity of the network.
The basic HDRP is enhanced by us by
combining the Back Propagation technique and
hence we also have simulation results for HDRP
(BP) which gave better performance. Our devised
protocol has been compared with Epidemic,
PRoPHET and Spray & Wait routing protocol in
DTN. These protocols are renamed as FixedEpi,
FixedPro and FixedSnW respectively as they are
applied to a network which has fixed infrastructure
that is fixed DTN routers.
Figure 4: Latency at different traffic intensity.
Figure 5 depicts that that the HDRP outperforms all
other protocol in terms of the end-to-end average
latency of the network. Whereas each of the
PRoPHET, SnW and Epidemic has latency in the
range of 700 sec the latency for the HDRP is less
than 50 sec. The reason behind HDRP performing so
much better than other protocols are mentioned
below:
The efficient Forwarding mechanism of
HDRP: In HDRP when a router has a message it
ACHIEVING LOW END-TO-END LATENCY WITH A HANDOFF-BASED DETERMINISTIC ROUTING
PROTOCOL (HDRP) IN DELAY-TOLERANT NETWORKING
17
looks for whether a connection is available: at
first the router tries whether a direct connection
can be found to the destination. If it is not there
then the router consults its Proxy List (PL) to
see if there is some information about the next
hop to reach the destination for the message. If
this second attempt also fails then the router
tries to send out the message by flooding it to
the available outgoing connections. The router
uses the flooding method as the last resort to
forward the message to its destination. In most
of the cases it can found a direct route or a
suitable next hop from its PL. Using HDRP a
router does not need to wait for a suitable
condition for forwarding to meet as in
PRoPHET a node with higher delivery
predictability has to be found or in a Spray and
Wait protocol the destination itself should be
there if the sparying phase is finished.
In our system model we consider the fixed
DTN routers interconnected with each other and
the end nodes, that is the FH and the MH are
communicating through this routers. During this
communication in between the fixed routers
there is no waiting time because of the fixed
links. The only waiting time is between the
Source and the router next and the Destination
and the router next. This logical time is also
very minimal. This contributes to the low end-
to-end latency of the network.
The existing DTN routing protocols deals
with only the mobile nodes and it is required to
wait until the next hop mobile node physically
moves to the wireless range of the previous hop
node and also the forwarding condition needs to
be satisfied before the routing takes place. This
leads to the larger value of overall end-to-end
latency.
HDRP (BP) gives slightly worse performance (avg.
38 sec.) than that of HDRP (avg. 32 sec.) because
of the fact that now more bundles can reach the
destination from the distant routers using the
cached route update information of the Back
Propagation technique.
Figure 5 also depicts how the other protocols will
perform in the same environment. It is found that
PRoPHET performs well when the traffic density is
low but the performance overall deteriorates as the
traffic density increases. Since this protocol works
based on finding a suitable node to forward with
higher delivery predictability than the previous one,
it can be concluded that increasing the traffic density
does not help much in finding that suitable next hop.
On the other hand the Spray and Wait (SnW)
protocol shows overall same performance for all
types of traffic density and the latency is larger than
the PRoPHET. Epidemic presents quite a low profile
than all other protocols.
Figure 6 presents the comparative situations of
the end-to-end latency for all the delivered messages
in the above mentioned protocols. This figure clearly
shows the lower latency range for the HDRP where
about 90% of the messages contribute between 5 to
15 sec latency whereas in case of PRoPHET and
SnW protocols most of the messages have latency
between 256 to 1025 sec range. This reveals that in
HDRP most of the messages are delivered to the
destination very quickly but in other protocols
majority of the messages takes longer time toreach
the destination and hence the end-to-end latency
increases. In this chart the HDRP (1, 5) gives more
end-to-end average latency than the HDRP (1, 500)
owing to the heavy traffic intensity which causes
more delay for the Bundles to reach the destinations.
We have plotted the PRoPHET (1,5) and SnW (1,5)
to show the comparison between the protocols when
the network traffic is very high.
Figure 5: Latency values of different number of messages.
4.2.2 Performance at Different Router
Density
We have simulated our network model for two
different topologies: i) Few Routers (FR):26 routers
ii) Many Routers (MR): 65. The routers are not fully
connected but they are well connected. They are
placed at different positions of the map of our model
so that the mobile nodes plying around the city map
can fall within the range of one or other to
accomplish their communication. It can be
intuitively guessed that increasing the router density
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
18
will certainly improve the delay performance in our
system. But we wanted to show how much would be
the difference if the number of the routers are more
than double.
Figure 7 depicts that HDRP and HDRP (BP)
have lower latency when Many Routers are present
than the Few Routers. This amount is substantially
lower than other protocols. As the number of router
increases more routers can be used to reach the
destination quickly. We found very low latency
values especially in case of MR. When the router
number increases from 26 to 65, the latency
decreases from 50sec (avg) to 25 sec (avg.) in case
of HDRP and from 60 sec (avg.) to 30 (avg.) sec in
case of HDRP (BP).
Figure 6: Latency values of different number of routers.
Another point worth mentioning here is the
transmitting range of the routers. We have used the
300 m transmitting range. With lower transmitting
range the performance degrades. And with higher
transmitting range we got same performances as
with the 300m range.
4.2.3 Performance for Different Node
Density
We wanted to show what will happen if the number
of end nodes communicating in our working
environment increases keeping the fixed router
numbers at 26. We tried with different end node
numbers from 124 to approximately 1000 nodes. For
the wide range of (120 to 1000 numbers of nodes),
both HDRP and HDRP (BP) gave much lower
latency value than others. Figure 8 shows that the
HDRP is scalable for the different node density.
4.2.4 Effect of Different Mobility Speed
To show how the mobility of the MHs affects our
protocol and others we have chosen different values
of Mobility speed, M by varying the speed of each
type of mobile nodes (pedestrians, cars, trams etc.),
0
100
200
300
400
500
600
700
800
0 200 400 600 800 1000 1200
Latency(sec)
Numberofnodes
HDRP FixedProphet
FixedEpidemic HDRP(BP)
Figure 7: Latency at different node density.
as explained in Table 1 For example, M1= [(.5, 1.5)
x40+ (2.7, 13.9) x40+ (7, 10) x2+ (7, 10)
x2]/84=4.833. Here the pedestrians speed has the
distribution of (.5, 1.5), no of pedestrians is 40;
speed distribution for the cars are (2.7, 13.9), no of
cars is also 40; trums1 and trums2 both have the
distribution as (7, 10) and the total number of mobile
nodes are 84. Finally, the total speed is divided by
the total number of nodes that is 84 and we get the
M1 value as 4.833.We varied the speed range of the
different type of mobile nodes and calculated the
M2, M3 and M4 respectively.
0
200
400
600
800
M1
(4.833)
M2
(7.8809)
M3
(18.2142)
M4
(27.261)
Laten cy(sec)
HDRP FixedPROPHET FixedEpidemic
Figure 8: Latency for different mobility speed.
Figure 9 shows that the latency of HDRP
decreases to about 50% of its value at M4 than M1.
As the moblity of the nodes increases from 4.833 to
27.261 that is more than 6 times, now more Bundles
can reach their destination faster than before. Thus
the latency of HDRP decreases to 21 sec from 43 sec
which is almost half. As we know that in Epidemic
protocol the message can spread faster in the
network if the speed of the mobile nodes increases
ACHIEVING LOW END-TO-END LATENCY WITH A HANDOFF-BASED DETERMINISTIC ROUTING
PROTOCOL (HDRP) IN DELAY-TOLERANT NETWORKING
19
and thus can reach the destination within small time.
So the latency for Epidemic decreases to 403 sec to
724 sec as mobility goes from M1 to M4. But
PRoPHET is less effeced by mobility than Epidemic
and ofcourse from HDRP as its operation is guided
by the suitable forwarding condition to be made.
4.2.5 Effect of Route-cache Time Variation
In the HDRP with Back Propagation technique we
varied the Alive time or route- cache time at each of
the routers from 5 sec to 3000 sec value range.
Because of the increased cache time at the routers,
few more Bundles get their way to the destination
deterministically using the cached route information.
The latency increases to 31.27 sec from 22.82 sec
which shows that now Bundles from the far away
routers are now contributing in the latency values
and so there is a increase in the overall latency.
5 RELATED WORK
5.1 Related Work on Handoff
Technologies in TCP/IP Protocol
There have been ample research works going on the
Mobility issues particularly Handover techniques in
Mobile Wireless environment. After wireless
networking technology became popular people tried
to develop mechanism that will deal with both the
wired and wireless part of a network efficiently.
Different methods were devised to overcome the
problems associated with the TCP to handle mobility
in the wireless environment (Manzoni et al., 1995), (
Yavatkar and Bhagawat, 1994), (Balakrishnan et
al.), (Caceres and Iftode, 1994). In spite of many
improvements these methods have the drawback of
end-to-end session management, TCP slow start
mechanism etc. I-TCP, Snoop-TCP, M-TCP and few
other protocols were developed to handle the
handoff situations efficiently through the use of
mediation by the Mobility Support Router (MSR)
(Bakre and Badrinath, 1997). But these methods also
suffer from the problem of large Handoff Latency
due to the connection states transfer between the old
and new MSR. In HDRP the rerouting during the
Handoff is done with the help of the hop-by-hop
reliability mechanism and custody transfer of the
DTN technology. This protocol does not have the
end-to-end session management or connection state
transfer problem during handoff. When handoff
takes place, the MH registers its location to the new
router and this location information is propagated
back to the experienced route and cached there so
that any Bundle destined to that MH can be
deterministically delivered to the destination.
The handoff latency is reduced in HDRP in
comparison to I-TCP in terms of number of message
exchanges:
Between MH and its CM it is similar because in
case of I-TCP we have Beacon/Greet/Grack and in
case of HDRP we have Beacon/REG/ SR
correspondingly. But between the CM and the PM
the number of message exchanges is not same
because I-TCP has Fwd Ptr/Fwd Ack/MHState/ACK
and HDRP has HO message/Data forwarding/SR.
Within the CM and PM routers, in case of I-TCP,
there are number of internal message exchanges
between the components of the router to accomplish
the handoff process but HDRP does not require any
internal message exchanges for the handoff to take
place.
Mobile IP, the mobility extension to the Internet
Protocol devises all the techniques to handle
mobility related and hence handover situations at the
network layer. But it also suffers from many
problems regarding the duration of handover and the
scalability of the registration procedure (Schiller). If
we consider a large number of mobile nodes
changing networks quite frequently, a high load on
the home agents as well as on the networks is
generated by registration and binding update
messages. The message delivery in HDRP does not
involve going through any home agent and update
messages do not need to travel so far. The old router
simply consults the PL and forwards the messages
destined for that particular mobile node. This
process is a very simple one and takes reasonably
less amount of time.
IP micro mobility-protocols like Cellular IP
(Campbell and Gomez-Castellanos, 2000),
(Campbell et al., 2000) or others are developed to
complement mobile IP by offering fast and almost
seamless handover control in limited geographical
areas. But they accompanies additional network
load induced by forwarding packets on multiple
paths. An additional cost of these schemes is that
communication, signaling and information state
exchange are required between the base stations for
these approach to work. On the other hand the
Handoff protocol in HDRP implements handover of
the messages with minimum number of control
message exchanges and no additional cost of
overhead. In case of Cellular IP, the back
propagation of route update packet takes place
between the MH and the crossover gateway of the
cellular network. In case of HDRP, the back
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
20
propagation takes place along the experienced route
of the MH that is through all the Previous Masters of
the MH. The distance between the MH and the
crossover Gateway is very important in case of
Cellular IP because while considering the Hard
handoff, the notification time from the new base
station to the old base station should be less than the
round trip time from the MH to the crossover
Gateway. This reduces the packet loss. On the other
hand, in HDRP (BP) the route update propagates
until sometime defined by the Alive time of the
route update Bundle at each of the router.
Another technique used in Cellular IP to reduce
the packet loss is the use of soft handoff technique
where the semisoft packet creates new routing cache
mappings between the crossover and the new base
stations, beforehand. In case of HDRP we didn’t use
any kind of route cache mapping creation before the
actual handoff takes place.
Many schemes have been developed considering
cross layer approach that is considering the Link
Layer triggering to the Network layer and how they
jointly reacts to handle the handover problems
(Blondia et al..,2003). In HDRP the handling of
both routing and handoff is done in a single bundle
layer which reduce cross layer interaction and
synchronization overhead and thus makes the
handoff latency smaller. An interaction between the
Bundle layer and the Physical layer is present during
the registration process by receiving the Beacon.
There are methods like Daelalus Implementation
(Seshan et al., 1996) that anticipates a handoff using
the received signal strength and multicast data
destined for the MH to nearby base stations in
advance. Combined with intelligent buffering
techniques at the base station, this provides good
performance without explicit data forwarding. But
this method has the inefficiency in handling the
routing of packets to the base station and the
overhead of buffering packets at several base
stations. During handoff in HDRP, we do not need
to consider any multicasting technique neither do we
need to apply any routing in advance. Hop by hop
routing decision is taken dynamically and buffering
is accomplished through custody transfer mechanism
efficiently.
5.2 Related Work on DTN Routing
Protocols
A number of routing protocols have been targeted
towards the context of intermittently connected
mobile networks with opportunistic connectivity.
Many of these protocols assume that all nodes are
mobile and have developed algorithms to transfer
message between these nodes. Flooding is one of the
popular techniques among these. Epidemic Routing
(Vahdat and Becker, 2000) is the protocol that
extends the concept of flooding in intermittently
connected mobile networks. It shows good
performance in a DTN environment where random
pair-wise exchanges of messages among MHs
ensure eventual message delivery and performs well
in terms of maximizing message delivery rate and
minimizing message latency until other protocols
were devised. PRoPHET (Lindgren and Schelen,
2003), a probabilistic routing protocol for such
networks assumes non-random mobility of nodes to
improve the delivery rate of messages while keeping
buffer usage and communication overhead at a low
level. The Spray and Wait routing protocol
(Spyropoulos and Raghavendra, 2005) manages to
significantly reduce the transmission overhead of
flooding-based schemes and have better
performance with respect to delivery delay
especially when the wireless channel has high
contention. These flooding based routing protocols
do not make use of the global knowledge and hence
suffers large latencies. We wanted to make use of
the knowledge of the location of the mobile node
and propose a handoff based routing protocol which
can route Bundles in a deterministic way. The route
update information during handoff and Back
Propagation and caching of this location information
over the experienced route improves the
performances.
In comparison to Epidemic, PRoPHET and
Spray & Wait protocols HDRP is deterministic in
nature and this is a more logic based routing
protocol. With carefully designed forwarding
mechanism and message prioritizing technique,
under different scenarios, it is possible to achieve
better delivery delay than the other protocols
mentioned above.
6 CONCLUSIONS
The HDRP protocol is a simple but efficient
handoff-oriented protocol that integrates the DTN
features with Custody Transfer and hop-by-hop
routing and the existing Internet-based handoff
schemes like mobile-IP, cellular-IP and I-TCP. This
integration is not a simple result that simply
combines the two different technologies but a
sophisticated and well-considered result because
DTN architecture is based on hop-by-hop routing
and fundamentally different from the end-to-end
ACHIEVING LOW END-TO-END LATENCY WITH A HANDOFF-BASED DETERMINISTIC ROUTING
PROTOCOL (HDRP) IN DELAY-TOLERANT NETWORKING
21
routing of the Internet architecture. Therefore we
have devised a unique integration that fully utilizes
the DTN features.
HDRP gives better performance than the already
existing DTN protocols in terms of Delivery ratio
and end-to-end latency by making use of the
location information of the MH and applying the
Back Propagation of the route update information
technique efficiently. The protocol shows superior
performance even for high traffic loads and it is
possible to achieve very low end-to-end latency with
the help of this protocol.
REFERENCES
Bakre V. and Badrinath B. R. (March 1997). Imple-
mentation and Performance Evaluation of Indirect
TCP, IEEE transaction on computer, vol.46, no.3.
Balakrishnan H., Seshan S. and Katz RH. (1995).
Improving reliable transport and handoff performance
in cellular wireless networks, Wireless Networks.
Blondia C., Casals O., Cerdà Ll., VAN D. W., Willems G.
and Cleyn P.De. (2003). Low Latency Handoff
Mechanisms and their implementation in an IEEE
802.11 Network, Proceedings of ITC18.
Caceres R. and Iftode L. (1994). The effects of mobility
on Reliable Transport Protocols, IEEE Journal of
selected areas in communication.
Campbell A.T. and Gomez-Castellanos J. (2000). IP micro
mobility protocols, SIGMOBILE Mobile
Computing and Communications Review.
Campbell A.T., Gomez J., Kim S., Valko A.G. and Wan
C. (August, 2000). Design, Implementation, and
Evaluation of Cellular IP, IEEE Personal
Communications.
Fall K. (Aug. 2003). A Delay-Tolerant Network
Architecture for Challenged Internets In ACM
SIGCOMM.
Keränen A., Ott J. and Kärkkäinen T. (March 2009). The
ONE Simulator for DTN Protocol Evaluation,
SIMUTools'09: 2nd International Conference on
Simulation Tools and Techniques. Rome.
Lindgren A., Doria A. and Schelen O. (2003).
Probabilistic routing in intermittently connected
networks, SIGMOBILE Mobile Computing and
Communications Review, 7(3):19-20.
Manzoni P., Ghosal D. and Serazzi G. (1995). Impact of
mobility on TCP/IP: an integrated performance study,
IEEE Journal on Selected Areas in Communications.
Ott J. and Kutscher D. (July 2005). Applying DTN to
Mobile Internet Access: An Experiment with HTTP,
Technical Report TR-TZI-050701, Universit¨at
Bremen.
Schiller J. Mobile communication, 2
nd
edition, Addison-
Wesley.
Scott K. and Burleigh S. (November, 2007). Bundle
Protocol Specification
Seshan S., Balakrishnan H. and Katz R.H. (1996).
Handoffs in Cellular Wireless Networks: The
Daedalus Implementation and Experience, Kluwer
International Journal on Wireless Communication
Systems.
Spyropoulos T., Psounis K. And Raghavendra C.S.
(August 22-26, 2005), Spray and Wait: An efficient
routing scheme for intermittently connected mobile
networks, SIGCOMM`05 Workshops.
The ONE: http://www.netlab.tkk.fi/tutkimus/d tn/theone
Vahdat A. and Becker D. (April,2000). Epidemic routing
for partially connected ad hoc networks, Technical
Report CS-200006, Duke University.
Yavatkar R. and Bhagawat N. (1994). Improving end-to-
end performance of TCP over Mobile Internetworks,
Mobile Computing Systems and Applications.
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
22