DALi: A Communication-Centric Data Abstraction Layer for
Energy-Constrained Devices in Mobile Sensor Networks
Christopher M. Sadler and Margaret Martonosi
Department of Electrical Engineering
Princeton University
{csadler, mrm}@princeton.edu
ABSTRACT
Communications in mobile and frequently-disconnected sen-
sor networks are characterized by low-bandwidth radios, un-
reliable links, and disproportionately high energy costs com-
pared to other system operations. Therefore, we must use
as efficiently as possible any periods of connectivity that we
have. For this reason, nodes in these networks need mech-
anisms that organize data to streamline search operations,
local computation, and communications.
This work proposes a Data Abstraction Layer (DALi),
which is inserted between the application layer and the file
system. DALi organizes data with networking in mind to
facilitate the development of services for Data Search, Nam-
ing, and Reduction that combine to make communications
more efficient. From the resulting two-tiered data hierarchy,
we develop a multi-layer drill-down search structure that
can locate data multiple orders of magnitude faster (and
with much lower energy) than simpler data storage struc-
tures. Additionally, DALi conserves energy and bandwidth
through a mechanism that acknowledges and removes spe-
cific data segments from a mobile sensor network. Finally,
it seamlessly integrates in a lossless compression algorithm
specifically designed for sensor networks to save additional
energy.
Categories and Subject Descriptors: C.2.1 [Network
Architecture and Design]: Wireless communication; C.3
[Special-Purpose and Application-Based Systems]: Real-
time and embedded systems; D.4.3 [File Systems Manage-
ment]: File Organization; E.5 [Files]: Sorting/Searching
General Terms: Algorithms, Management, Performance
Keywords: Data Search and Storage, Energy Efficient
Communications, Mobile Ad Hoc Sensor Networks
1. INTRODUCTION
Mobile and frequently disconnected sensor networks form
an interesting subset of the sensor network design space.
The target applications vary drastically, from zebra track-
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
MobiSys’07, June 11-14, 2007, San Juan, Puerto Rico, USA.
Copyright 2007 ACM 978-1-59593-614-1/07/0006 ...$5.00.
ing [39] to polar monitoring [4]. They have an underlying
set of common traits, however, largely based on their harsh
operating environments which can make physical node ac-
cess difficult and which place logistical limits on the size of
the deployments.
These systems are characterized by both the severe re-
source constraints of sensor nodes and by short periods of
unreliable, low quality communications over low bandwidth
radios. Beyond their sensors, they collect and store data us-
ing an ultra-low power microcontroller and energy-efficient,
non-volatile memory in an effort to operate for months at a
time on a limited energy budget.
Over time, sensor node microcontrollers have become
more capable, the amount of storage space has increased,
and the energy costs of CPU and storage have decreased.
These trends are likely to continue. However, radio trans-
missions have remained expensive and unreliable and this is
unlikely to improve significantly over time. Significant chal-
lenges exist regarding the physical energy costs of wireless
signal propagation, the difficulties of designing appropriate
antennas, and environmental factors which are exacerbated
by a constantly changing network topology. Additionally, in
a mobile network, nodes may transmit multiple replicated
copies of the data to balance latency and energy constraints
[32][34]; unnecessarily transmitting the data either to nodes
that already have it or to anyone after the sink has received
a copy wastes valuable bandwidth and energy.
As a result, a good mobile sensor system must be designed
with the data storage and communication infrastructure in
mind. Current Flash file systems designed for stationary
sensor networks offer clear advantages over raw application
management of data, but on their own these systems do not
meet our goals. For example, files can grow to the size of
the Flash, there is no efficient way to identify particular data
items in files, and there is minimal support for compression.
However, for the tasks for which they were intended, such as
using the Flash efficiently and ensuring data integrity, these
file systems perform well. For this reason, we have developed
DALi, a Data Abstraction Layer for mobile sensor networks
that lies between the application and the file system and
provides nodes with Data Search, Naming, and Reduction
services.
Data Search is the ability to quickly locate specific data on
the node, by name or by value, and summarize it when ap-
propriate. We emphasize search speed because minimizing
query response times improves bandwidth efficiency.
Data Naming is the ability to identify specific sections of
data in a granularity that can be easily transmitted through
99

the network. Using this, we can build “delete lists” which
aim to conserve energy and bandwidth by stopping data
(once delivered to the sink) from being unnecessarily trans-
mitted further. We can likewise prevent nodes from trans-
mitting data to other nodes that already have it.
Data Reduction is the ability to shrink the data through
in-network computation, data aggregation, or compression.
This mechanism conserves energy and bandwidth by reduc-
ing the volume of data in the network.
These three functions are interdependent and a truly ef-
fective system for mobile sensor networks needs to provide
services for all three.
The contributions of this work include:
• We design and develop a prototype of a Data Abstraction
Layer (DALi) that restructures data in a way that sim-
plifies communications and uniquely incorporates each of
the processes of Data Search, Naming, and Reduction.
• We introduce an efficient way to incorporate “delete
lists” into the system, which can reduce energy con-
sumption by multiple orders of magnitude by reducing
unnecessary transmissions.
• We demonstrate that our hierarchical data organization
serves as the basis for a drill-down search structure that
allows for simple, fast sensor data searches on both spa-
tial and temporal data. DALi can effectively search large
real-world datasets in the amount of time it takes to send
a handful of packets.
The remainder of this paper is structured as follows. Sec-
tion 2 discusses related work. Section 3 then presents an
overview of DALi. Section 4 introduces DALi’s Data Search
service and Section 5 introduces its Data Naming and Data
Reduction services. Then, Section 6 offers a discussion and
evaluation of our work, and Section 7 concludes the paper.
2. RELATED WORK
Sensor network data management systems are typically
tailored to stationary, well-connected sensor networks and,
therefore, do not attempt to leverage Flash data storage to
make opportunistic mobile communications more efficient.
Additionally, no existing data organization offers a combina-
tion of Data Search, Naming, and Reduction services similar
to those offered by DALi.
Proactive relational query processors [11, 13, 19, 38] use
queries to activate specific sensors on a node, collect read-
ings for a given period of time, and return the results. They
may also compute data summaries and aggregate data from
multiple nodes. However, mobile delay-tolerant networks
must be reactive because of the long latency required to de-
liver queries to nodes. Additionally, the often-changing net-
work topology prevents nodes from employing aggregation
techniques that rely on data correlation in the network or
distributed schemes that rely on specific sensors to execute
specialized data reduction algorithms; nodes should primar-
ily rely on data reduction algorithms which can be executed
locally. For stationary sensor networks, both more tradi-
tional [30] and distributed [24] storage abstractions exist as
well, but frequent disconnections and the sparse distribution
of nodes prohibit us from using these methods.
Generic Flash file systems [1, 35] and Flash file systems
for sensor networks [6, 7, 10] store data in arbitrarily large
files, like PC-based file systems. These files can be far larger
than mobile nodes can transmit in one communications pe-
riod. The file systems have no mechanisms for identifying
smaller segments of the file, which is critical to preventing
unnecessary communications related to duplicate copies of
data in the network. Our work provides the additional ser-
vices necessary for proper data location and identification
to assist communications.
Our mechanism for subdividing data into smaller chunks
is similar to the BitTorrent peer-to-peer file distribution sys-
tem [5]. However, existing variants for MANETs [23] and
sensors [33] are inappropriate for our networks because each
node uses a tracker to find out which peers have the file
it wants—information that is not likely to be available—
and assume good connectivity and reliable multi-hop routes
through the network that can move large volumes of data
at once. DALi, on the other hand, gradually acquires data
over the independent, opportunistic peer-to-peer links char-
acteristic of mobile sensors.
Our data division mechanism also resembles the SPIN
routing protocol [12, 15], which, in simulation, breaks data
into 500B segments and uniquely names them in an effort to
suppress redundant transmissions. However, SPIN requires
that each application provide its own naming scheme. DALi,
on the other hand, provides a standard two-level naming
structure which is applicable across applications and allows
names to be merged so that one name can represent much
more data. It also provides search and data reduction ser-
vices that SPIN does not consider.
Other Flash-based sensor storage systems offer useful data
structures (e.g., Capsule [20]) and search capabilities (e.g.,
MicroHash [16]) and are a strong influence for our work.
However, they do not attempt to tackle the issue of sim-
plifying communications and they were never intended to
transmit more than short data summaries, directly related
to their intended use on stationary, connected networks. Ad-
ditionally, the search algorithms for systems such as Micro-
Hash are only designed to handle data collected from a single
node and will not work if the data is not stored in time or-
der. Unordered data is common in systems that store and
process data from multiple nodes like DALi does.
Ganesan et. al. use wavelet summarization both on a sin-
gle node and over groups of nodes to offer multiple granulari-
ties of data for transmission and search [9]. Their concept of
drill-down queries is similar to ours, but we generate meta-
data rather than wavelets since they are not appropriate for
answering queries on all types of data. Additionally, we can-
not expect to have enough nodes, the proper node topology,
or the data correlation necessary for wavelets to be effective
across groups of nodes.
Delay-Tolerant Networks [8] may use data mules to gather
data files from stationary sensor nodes [26]. However, com-
munications between the sensing nodes and the data mule
may be unpredictable, unreliable, and intermittent, espe-
cially if the mule’s movements are random. DALi can assist
data delivery in these scenarios by dividing files into more
communicable segments.
3. DALI ARCHITECTURE
Nodes in sparse, mobile sensor networks will often adapt
between either sending all of their data, which minimizes
latency, or responding to specific queries, which minimizes
communications. If the application wants all nodes to send
all data, nodes need intelligent ways to prevent costly un-
necessary communications and to improve the efficiency of
100