Content uploaded by Leslie Fife
Author content
All content in this area was uploaded by Leslie Fife on Sep 03, 2014
Content may be subject to copyright.
Research Issues for Data Communication in Mobile Ad-Hoc Network Database Systems
Leslie D. Fife, Le Gruenwald
*
Brigham Young University – Hawaii, Computer Science Department, Laie, HI 96762
*
The University of Oklahoma, School of Computer Science, Norman, OK 73019
Email: ldfife@cs.byuh.edu
,
*
ggruenwald@ou.edu
Abstract
Mobile Ad-hoc Networks (MANET) is an emerging
area of research. Most current work is centered on
routing issues. This paper discusses the issues
associated with data communication with MANET
database systems. While data push and data pull
methods have been previously addressed in mobile
networks, the proposed methods do not handle the
unique requirements associated with MANET. Unlike
traditional mobile networks, all nodes within the
MANET are mobile and battery powered. Existing
wireless algorithms and protocols are insufficient
primarily because they do not consider the mobility
and power requirements of both clients and servers.
This paper will present some of the critical tasks
facing this research.
1. Introduction
A traditional mobile network consists of a fixed
network of servers and clients, with a collection of
mobile clients that move throughout the geographic
area of the network. Within the mobile network,
servers have unlimited power and communicate with
mobile hosts over a wireless connection. Mobile
clients may only communicate among themselves
through a server. Among the issues in this type of
network are client power consumption, connectivity
of the network, and reachability of mobile clients
from a server.
In contrast, a MANET is a collection of mobile
servers and clients. All nodes are wireless, mobile
and battery powered [9]. The topology can change
frequently. The nodes organize themselves
automatically, and can be a standalone network or
attached to a larger network, including the Internet
[2]. All nodes can freely communicate with every
other node.
In addition to the issues associated with a mobile
network, the power consumption and mobility of the
server(s) must also be considered in a MANET.
Originally called Mobile Packet Radio, Mobile
Ad-hoc Network (MANET) technology has been an
important military research area [5]. This technology
has practical use whenever a temporary network with
no fixed infrastructure is needed. Other uses include
rescue operations and sensor networks [13][18]. The
support of these military and civilian uses often
requires the presence of a database to store and
transmit critical mission information such as
inventories and tactical information.
There is one other crucial characteristic of a
MANET. Traditional mobile networks involve the
server in all data communication. MANET includes
the traditional database capabilities of data push and
data pull, but it also allows the clients to
communicate directly with each other without the
involvement of the server, unless necessary for
routing [6][13].
2. MANET Architecture
The nodes in a MANET can be classified by
their capabilities. A Client or Small Mobile Host
(SMH) is a node with reduced processing, storage,
communication, and power resources. A Server or
Large Mobile Host (LMH) is a node having a larger
share of resources [9]. Servers, due to their larger
capacity contain the complete DBMS and bear
primary responsibility for data broadcast and
satisfying client queries. Clients typically have
sufficient resources to cache portions of the database
as well as storing some DBMS query and processing
modules [9].
As both clients and servers are mobile, the speed
at which the network topology changes can be rapid.
A variety of techniques have been proposed to assist
in the routing tasks of MANET. New protocols were
necessary as the protocols for fixed infrastructures
and static networks do not perform well when node
mobility is included [16]. A global routing structure
is also not useful in MANET due to its dynamic
topology and need for distributed control [16]. Work
on routing is ongoing and is coordinated through the
Internet Engineering Task Force (IETF) [3].
MANET characteristics include a preference for
reactive (on-demand) routing, unpredictable and
frequent topology changes and distributed control
[18]. The primary MANET limitations remain limited
bandwidth and battery power [18].
Nodes may not remain connected to the network
throughout their life. To be connected to the
network, a node must be within the area of influence
of at least one other node on the network and have
sufficient power to function.
In Figure 1, a few nodes of a MANET are shown
graphically. It is important to note that each node has
an area of influence. This is the area over which its
transmissions can be heard. A LMH will initially
have a larger area of influence as it generally has a
more powerful battery. As the power level decreases,
the area of influence of any node will shrink. This is
due to the fact that the power available to broadcast is
reduced.
Network nodes may operate in any of three
modes that are designed to facilitate the reduction in
power used [12][19]:
• Active Mode (or Transmit Mode)
: this is the
mode using the most power. It allows both the
transmission and reception of messages and
consumes 3000 to 3400 mW [19].
• Doze Mode (or Receive Mode)
: the CPU is
capable of processing information and is also
capable of receiving notification of messages
from other nodes and listening to broadcasts.
1500 to 1700 mW are consumed in this mode
[19].
• Sleep Mode (or Standby Mode)
: the CPU does
no processing and the node has no ability to
send/receive messages. The node is inactive and
consumes only 150 to 170 mW [19]. This mode
allows a node to turn itself off for short periods
of time without requiring power-up or re-
initialization.
A node with no remaining power, or one that is
off, is not currently a part of the network.
It is clear from the description and Figure 1 that
a node may not be reachable by another node (LMH
or SMH). In other words, nodes may become
disconnected from the entire network. When moving
back in range of other nodes, they will become re-
connected. Conversely, a node may be reachable by
several LMHs or SMHs. The potentially rapid and
regular reconfiguration of the network topology is
routine with the MANET.
3. Current Data Communication Research in
Mobile Databases
Data communication research in mobile
databases is limited to situations where only clients
are mobile and battery powered. These studies are
concerned with ways to maximize client battery life
by improving either the organization of the data
broadcast or the selection of the broadcast contents.
Below is a brief discussion of representative papers
in this area.
Aksoy, et. al. [4] present a large-scale on-
demand broadcast model called RxW (Requests times
Wait). At each broadcast tick, the server chooses an
item to broadcast based on the number of request and
the amount of time the original request has been
waiting. The overhead for large databases is
significant in both time and space [9]. In addition,
the server may be constantly in active mode, as the
server power level is not an issue. There is a power
cost associated with constant client queries.
Guo, et. al. [10] also work on improving the
responsiveness of database service. In their
approach, the server maintains a list of popular and
less popular items. The popular items are
continuously broadcast. If a less popular item is
needed, a client may request it. This interrupts the
broadcast, which continues with the data broadcast
after serving the request. The server never stops
broadcasting, consuming power.
Yajima, et. al. [22] and Grassi [8] approach the
problem differently. They try to improve database
service by the organization and use of the broadcast.
Yajima [22] builds broadcasts where highly
correlated items are found together in the broadcast,
minimizing the number of times a client must access
the broadcast. Grassi [8] uses prefetching related
items into the client cache so that they will be
available locally if needed. While prefetching may
shorten the time a client needs to access a data item,
prefetching wastes power and space through
accessing and storing broadcast items that may not be
needed. The benefits from a correlated broadcast
require a constant processing and broadcasting by the
server, leaving it constantly in active mode.
While addressing the issues of broadcast size,
organization and content, the traditional mobile
broadcast methods proposed fail to deal with server
mobility and power limitations.
4. Current Data Communication Research in
Mobile Ad-Hoc Databases
A MANET may include data pull, data push and
peer-to-peer communication. No research has been
done which includes all three forms of
communication. However, data push and data pull
have been addressed to varying degrees. Below the
Figure 1 – MANET Architecture
SMH -
Client
LMH Area of
Influence
LMH -
Server
SMH Area of
Influence
recent work in Mobile Ad-Hoc data communication
is addressed.
Wieselthier, et. al. have been working together
on MANET broadcast issues. Their approach is the
construction of a minimum-energy tree rooted at the
broadcast source [20][21]. Two algorithms called
Broadcast Incremental Power (BIP) and Multicast
Incremental Power (MIP) have been advanced for
building these trees. The BIP builds the minimum
energy tree for a broadcast, while the MIP uses the
BIP algorithm, but only includes those branches
necessary to reach the clients needing to receive a
specific broadcast [20].
The algorithms were tested and showed that by
utilizing broadcast in a mobile environment, energy
savings can be achieved. Further studies with larger
networks were recommended [20]. However, node
mobility was not addressed.
The cost of building the tree is considered
negligible by the authors as the use of the tree is long
when compared to the building of the tree [20]. This
would be the case for stationary nodes. However,
stationary nodes would be the exception in MANET.
They accommodate “movement” with the
observation that increasing transmitter power will
allow them to reach nodes in new locations [21]. No
potential interference between broadcasts and no
need to rebuild the tree once created are considered.
The restrictions and assumptions are limiting. In
addition, tree-based protocols do poorly with node
mobility [10]. The problems of limited bandwidth,
the need for tree maintenance, and node mobility
remain.
Two algorithms to handle data push and data pull
within the MANET were proposed in [9]. The first is
the adaptive broadcast scheduling algorithm. Within
this algorithm there are two potential ways to
construct a broadcast. New items may be either
added to the algorithm or may replace less important
data items [9].
A global network where all servers in a region
know the location and power of all other servers in
the region and full replication of the database is
assumed. Periodically, each server broadcasts its
location and power level. This begins the broadcast
cycle [9]. This is a soft real-time system. There are
deadlines for data delivery. The deadlines were used
to determine which data request to service although
no penalty for missing a deadline was mentioned.
There is a leader protocol that selects the server
in a region with the greatest remaining power. The
leader coordinates the broadcast responsibilities of
other servers in its area of influence [9]. The lead
server determines which portion of a broadcast each
server transmits. The power level of each server
drives this broadcast assignment. The server with the
least power transmitted the most important data items
[9]. No server transmits the entire broadcast unless it
is the only server in a region
After the conclusion of broadcasting, clients are
permitted to query the servers. After the time period
for queries, the broadcast cycle repeats [9].
This initial algorithm has a potentially large
communication overhead, servers with no clients still
broadcast, and less popular items may starve or be
broadcast too late [9].
The second algorithm utilizes a popularity factor
(PF), as suggested by Datta et. al. [7]. The PF is a
measure of the importance of a data item. The PF
increases each time a request is made for a data item
[9]. The amount of time since the request was made
also affects the PF. If it has been too long, the need
to broadcast the item may be gone. This factor is
called the Resident Latency (RL) and is system and
scenario specific [9]. The PF decreases whenever a
request exceeds the RL value [9]. The PF is used to
assist in the building of relevant broadcasts and
includes RL in order to make allowances for the
movement of nodes. When the PF of broadcast items
is high, the probability of a broadcast that serves
maximum needs increases.
If a server has not received any requests for a
certain number of broadcasts, it will sleep rather than
broadcast to an empty audience [9]. Finally, to
localize data delivery, the lead server assigns each
server the amount of data to broadcast but not the
items to broadcast [9].
To deal with insufficient power levels, the
servers rebroadcast the previous index and broadcast
if they have insufficient power to build a new
broadcast [9]. It is not clear why broadcasting old
information is preferable to no broadcast at all.
This approach is still not sufficient as servers can
be assigned a broadcast larger than their power levels
would permit. Power and bandwidth is also wasted
with duplication.
5. Research Issues
The data communication research issues in
MANET databases center around two areas. The first
area concerns the limitations of the environment
(wireless, limited bandwidth, battery powered). The
second area concerns the many ways in which data
communication may take place.
Environmental Limitations
Power Consumption.
Power consumption is a concern in any mobile
network. However, in traditional mobile networks,
only the power needs of the clients are considered.
Here, the power of the server, which provides DBMS
data services, is perhaps more important as it
provides DBMS services to potentially many clients.
This is the one overriding issue [17]. Its parts are:
• Are power settings broadcast for servers and
clients? If so, how often?
• How do server power levels affect broadcast
assignments?
• What should be done with a LMH/SMH with a
low power level.
A server’s power setting is an important input
into the entire process. Servers with the greatest
power availability may be expected to perform the
most work. If this information is broadcast, power is
consumed.
Methods to minimize client power drain are also
important but are addressed in existing mobile
network research.
Broadcasting is both time and energy conscious
[15]. A carefully coordinated set of broadcasts can
reach a large number of clients who only have to
listen to get the information they need. Only if the
information needed is not broadcast does a client
need to query the database.
Timing
Regardless of the method of communication
used, access time and tuning time must be
considered. Tuning time is the measure of the
amount of time each node spends in Active Mode.
This is the time of maximum power consumption for
a client. Because of that, tuning time minimization is
an important goal.
Access time measures the responsiveness of the
algorithm. Access time refers to the amount of time a
client must wait to receive an answer to a database
query. If Access Time is too long, the client may no
longer be reachable from the server assigned to
broadcast the information. Power is wasted if
broadcast information is missed and must be
requested.
Data integrity
With data integrity, we are concerned with the
accuracy of information stored at each node: server
and client alike. This problem occurs as servers and
clients move in and out of contact.
Acknowledgement messages are not appropriate
in a MANET as mobility makes receipt unreliable
and extra bandwidth and power are consumed.
Data replication is an important consideration. If
the database is fully replicated among all mobile
servers, additional power is consumed to maintain the
databases. If full replication is not required or
possible, other data integrity issues exist.
While data replication may not exist for an entire
network, it may be possible to maintain it in disjoint
partitions within the network. Partitioning of a
network or database is either carefully designed and
reasonably static or is considered a failure condition
[14]. As servers are mobile, partitions would be
necessarily dynamic. Partitioning would also not be
considered a failure in a MANET, but would be
normal. Either method or level replication would add
some amount of overhead to the network.
The end result is that the database stored at each
server may not be consistent with one another. As
database updates are made, not all servers are
guaranteed to receive the updates in a timely fashion.
Nodes become disconnected for a variety of
reasons. This may be due to location or lack of
power. The dynamic nature of MANET makes
maintaining the data a challenge. Multiple versions of
the same information may exist throughout the
network. When portions of the network become
separated for a time, keeping data accurate may
become impossible.
Data Broadcast (Data Push)
Of all the MANET activities, data
communication remains one of the high power
consumption activities. When broadcasting, each
node listening to a broadcast consumes nearly half as
much power as is consumed by the broadcaster [12].
Traditional mobile network protocols [1][4]
assume that the clients can regularly submit requests
to the DBMS servers. Traditional methods [4] also
use frequency of request when building broadcasts.
There is nothing efficient about multiple clients
individually requesting the same data item. It is also
not energy efficient for servers to unicast the same
data individually to several clients. It is important to
minimize data requests, saving power at both the
server and client. Current methods do not minimize
client requests.
It is important to keep in mind that while
broadcast is energy efficient when working with
multiple nodes; it is not sufficient when a large
number of data items must be delivered [10]. Data
pull alone is also not sufficient [10]. Both methods,
used appropriately, are necessary to achieve the
greatest energy efficiency [10].
Broadcast Content
The size and contents of a broadcast affect power
consumption and the frequency of data queries. If the
broadcast is too large, unnecessary information may
be broadcast. If too little information or the wrong
information is broadcast, on-demand requests
increase. In both cases, Access Time also increases.
Traditional mobile networks solve this problem
through building broadcasts based on frequency of
queries [10] wasting client power or by broadcasting
continually [4] wasting server power.
Mobile network research shows that an index can
minimize the amount of time clients must remain
active, accessing the broadcast [11]. The tradeoff is
that the index must also be broadcast. The small
amount of energy needed to broadcast the index may
offset the large amount of energy needed by many
clients to listen to the entire broadcast.
• How often the contents of a broadcast are
built/changed.
• Node’s data needs – as determined from data
requests not served through data-on-demand or
peer-to-peer communication.
• What criteria are used to determine what is
included in the broadcast?
• Should an index be used as part of the data
content?
Broadcast Allocation.
If multiple servers exist in an area, who
broadcasts what? The methods proposed in [9]
assume a leader that coordinates the work of the
server group. This is an attempt to save power by
sharing the load. But must a leader be selected?
Perhaps each server can coordinate based on
individual knowledge of area servers and clients.
In addition to the allocation of broadcast content,
the timing of broadcasts is critical. In many ways,
MANET broadcasting is like the telephone party line
or a bus network topology. If several servers attempt
to broadcast simultaneously, there will be a collision
and the broadcast of all will be garbled. This is a
waste of time and power for both the servers and the
clients listening to the broadcasts. If using a lead
server makes broadcasts assignments, it is possible
for a node’s assignment to be larger than it can
accommodate, based on the node’s remaining power.
This is not an efficient allocation. A portion of the
broadcast will not be sent and that LMH will
disappear from the network due to running out of
power.
Broadcast Frequency
Too frequent broadcasts waste power
unnecessarily. Too infrequent broadcasts lead to
increased client requests, wasting their power. The
frequency of broadcasts will be a function of server
power levels and the data request frequency of
clients. Frequency of broadcasts affects both Tuning
Time and Access Time.
Broadcast Reasonableness
This is a question of whether or not to even
broadcast. If no clients are in the broadcast area, a
broadcast is meaningless and a waste of power. If
only a few are in the area of influence, handling data
needs interactively may be more efficient. A method
to identify and track nodes in a server's area of
influence is necessary.
Data-on-Demand (Data Pull)
Should a request be added to the next broadcast
or served immediately?
• Should a client be prohibited from querying for
the same data as another client in the same area
or should it just wait for data service?
• Does the server need to know how many clients
want a piece of data to determine data
importance?
• How is data aged so that all requested data is
eventually broadcast?
• Is it important to serve data requests even after a
certain amount of time has elapsed?
• When a SMH leaves an area, do we forward the
data service request – or do we rely on the SMH
to determine it is in a new cell and know it must
re-request the data?
• How do we forward service requests in the
network?
The issues here center on client data needs that
are not met by the data broadcast. If the broadcast
does not satisfy the needs of a client it must obtain
the data from a server (data-on-demand) or from
another client (peer-to-peer).
Which method a client uses will depend on who
has the most recent data. What factors determine the
best data source must be investigated. While
satisfying the data needs of the client, we must
remain sensitive to power consumption and mobility
issues
Peer-to-Peer Communication
When the server does not satisfy the data needs
of the SMH through broadcast or data-on-demand, a
client may communicate directly with another client
that has the needed data. The issues here include the
role of the different nodes in this communication, and
determining who has the data. Existing data
broadcast algorithms proposed in the MANET
literature do not address peer-to-peer communication.
• Should the client be limited in the number of
peer-to-peer requests they make?
• How does the server know it needs to route a
request?
• Should peer-to-peer be limited to certain types of
requests?
• If a request is not serviced in time, should it be
added to the next broadcast?
5. Conclusions and Future Work
Data communication is an important topic that
needs to be addressed when designing database
systems in MANET environments. This topic
involves far more than network routing. In addition,
existing mobile protocols are insufficient. They are
not geared towards the specialized needs of a
MANET.
The areas of concern within MANET data
communication are raised. Future research will need
to begin to address these issues. Along with these
issues, standardized benchmarks and criteria for
evaluation must be established so that proposed
protocols and methods can be legitimately compared.
6. Bibliography
[1] Mobile Ad-hoc Networks Working Group 2002,
www.ietf.org/html.charters/manet-charter.html.
[2] Section 2.5.3. In
Proc. 54th Internet Engineering
Task Force
July, 2002.
[3] The Internet Engineering Task Force (IETF).
www.ietf.org,
2002.
[4] Aksoy, D. and Franklin, M. Scheduling for Large-
Scale On-Demand Data Broadcasting. In
Proc.
12th International Conf. on Information
Networking
pp. 651-659, January, 1998.
[5] Corson, M., Freebergyser, J., and Sastry, A.,
"Mobile Ad Hoc Networking: Editorial,"
Mobile
Networks and Applications
, 4(3): pp. 137-138,
1999.
[6] Das, S., Castañeda, R., and Yan, J., "Simulation-
Based Performance Evaluation of Routing
Protocols for Mobile Ad Hoc Networks,"
Mobile
Networks and Applications
, 5(3): pp. 179-189,
2000.
[7] Datta, A., VanderMeer, D., Kim, J., Celik, A., and
Kumar, V. Adaptive Broadcast Protocols to
Support Efficient and Energy Conserving
Retrieval from Databases in Mobile Computing
Environments. A TimeCenter Technical Report
(University of Arizona), 1977.
[8] Grassi, V. Prefetching Policies for Energy Saving
and Latency Reduction in a Wireless Broadcast
Delivery System. In
Proc. 3rd ACM International
Workshop on Modeling, Analysis and Simulation
of Wireless and Mobile Systems (MSWiM)
, pp.
77-84, 2000.
[9] Gruenwald, L., Javed, M., and Gu, M. Energy-
Efficient Data Broadcasting in Mobile Ad-Hoc
Networks. In
Proc. International Database
Engineering and Applications Symposium
(IDEAS '02)
, July, 2002.
[10] Guo, Y., Pinotti, M., and Das, S., "A New Hybrid
Broadcast Scheduling Algorithm for Assumetric
Communication Systems,"
ACM Mobile
Computing and Communications Review
, 5(4):
pp. 39-54, 2001.
[11] Imielinski, I., Viswanathan, S., and Badrinath, B.
Energy Efficient Indexing on Air. In
Proc. ACM
SIGMOD International Conf. on Management of
Data (SIGMOD 94)
, pp. 25-36, May, 1994.
[12] Jones, C., Sivalingam, K., Agrawal, P., and
Chen, J., "A Survey of Energy Efficient Protocols
for Wireless Networks,"
Wireless Networks
, 7 pp.
343-358, 2001.
[13] Kahn, J., Katz, R., and Pister, K. Next Century
Challenges: Mobile Networking for "Smart Dust".
In
Proc. 5th International Conf. on Mobile
Computing and Networking (MOBICOM '99)
, pp.
271-276, August, 1999.
[14] Keum, C., Choi, W., Hong, E., Kim, W., and
Whang, K. Performance Evaluation of Replica
Control Algorithms in a Locally Distributed
Database System. In
Proc. 4th International
Conf. on Database Systems for Advanced
Applications (DASFAA 95)
, pp. 388-396, April,
1995.
[15] Lam, K., Chan, E., and Yuen, J. Broadcast
Strategies to Maintain Cached Data for Mobile
Computing Systems. In
Proc. Advances in
Database Technologies - LNCS 1552 (ER '98)
,
pp. 193-204, November, 1998.
[16] Lee, S., Su, W., and Gerla, M., "Wireless Ad Hoc
Multicast Routing with Mobility Prediction,"
Mobile
Networks and Applications
, 6(4): pp. 351-360,
2001.
[17] Lettieri, P., Fragouli, C., and Srivastava, M. Low
Power Error Control for Wireless Links. In
Proc.
3rd International Conf. on Mobile Computing and
Networking (MOBICOM '97)
, pp. 139-150,
September, 1997.
[18] Liu, J., Zhang, Q., Li, B., Zhu, W., and Zhang, J.,
"A Unified Framework for Resource Discovery
and QoS-Aware Provider Selection in Ad Hoc
Networks,"
ACM Mobile Computing and
Communications Review
, 6(1): pp. 13-21, 2002.
[19] Singh, S., Woo, M., and Raghavendra, C. Power
Aware Routing in Mobile Ad Hoc Networks. In
Proc. 4th International Conf. on Mobile
Computing and Networking (MOBICOM '98)
, pp.
181-190, October, 1998.
[20] Wieselthier, J., Nguyen, G., and Ephremides, A.,
"Algorithms for Energy-Efficient Multicasting in
Static Ad Hoc Wireless Networks,"
Mobile
Networks and Applications
, 6(4): pp. 251-263,
2001.
[21] Wieselthier, J., Nguyen, G., and Ephremides, A.
Resource-Limited Energy-Efficient Wireless
Multicast of Session Traffic. In
Proc. 34th Hawaii
International Conf. on System Sciences (HICSS-
34)
, January, 2001.
[22] Yajima, E., Hara, T., Tsukamoto, M., and Nishio,
S. Scheduling and Caching Strategies for
Broadcasting Correlated Data. In
Proc. 16th ACM
Symposium on Applied Computing
pp. 504-509,
March, 2001.