Sunteți pe pagina 1din 7

Proceedings of iiWAS2009 iiWAS 2009 Short Papers

Combining Cloud Computing and Wireless Sensor


Networks
Werner Kurschl Wolfgang Beer
Upper Austria University of Applied Sciences Software Competence Center Hagenberg
Softwarepark 11 Softwarepark 21
A-4232 Hagenberg, Austria A-4232 Hagenberg, Austria
werner.kurschl@fh-hagenberg.at wolfgang.beer@scch.at

ABSTRACT the IT hardware and system software infrastructure, as well as the


Just recently emerged a new paradigm for internet-based software applications delivered though this new service infrastructure.
systems which is called as cloud computing. The cloud provides Applications are delivered as a service over the Internet. These
scalable processing power and several kinds of connectable services or applications are generally known and defined as SaaS,
services. This distributed architecture has many similarities with which is also used here in context of Cloud Computing in the
a typical wireless sensor network, where a lot of motes, which are same manner. We will use the term Cloud as the hardware and
responsible for sensing and local preprocessing, are system software within a datacenter that provides the
interconnected with wireless connections. Since wireless sensor infrastructural background to host SaaS.
networks are limited in their processing power, battery life and Figure 1 shows that the Cloud provider has the responsibility to
communication speed, cloud computing usually offers the offer reliable hardware resources and necessary system software
opposite, which makes it attractive for long term observations, that allows a transparent hosting of services for the SaaS
analysis and use in different kinds of environments and projects, providers. It is important that the architectural design of SaaS
since the basic infrastructure remains the same. services has to consider new billing models that support the Cloud
Computing pay-by-use value chain. The provision of SaaS
In this paper we present a model, which combines the concept of
services most of the time is given through Web applications or
wireless sensor networks with the cloud computing paradigm, and
directly through Web Services.
show how both can benefit from this combination.

Categories and Subject Descriptors


D.2.11 [Software Architectures]: Patterns; C.2.4 [Distributed
Systems]: Distributed applications;

General Terms
Distributed System, Performance, Design, Reliability, Scalability

Keywords
Wireless Sensor Networks, Cloud Computing, Monitoring, Data
Analysis, Web Application

1. INTRODUCTION
Just recently emerged a new paradigm for internet-based software
systems which is called as Cloud Computing, which is to some Figure 1. Cloud Computing and SaaS
extent much driven by prominent software providers like Google
or Microsoft. The Cloud Computing paradigm is based on existing ideas like
Utility Computing, SaaS and Web Services and adds some new
Cloud Computing is often referred to as the next step beyond the aspects. According to the work from [1], three new hardware
Software as a Service (SaaS) paradigm, but there exists no general related aspects are introduced by the Cloud Computing paradigm:
accepted definition of Cloud Computing. In an excellent technical
report [1] written by a group of researchers from Berkeley 1. The illusion and transparent use of an unlimited amount
university an attempt for a general definition and scope of Cloud of resources that can be used by SaaS. Therefore
Computing is given. It shows that Cloud Computing both include application or service provider do not have the need to
plan the amount of resources in advance, instead they
dynamically scale the resources on demand.
Permission to make digital or hard copies of all or part of this work for 2. System software has to be aware that services are
personal or classroom use is granted without fee provided that copies are hosted according to a pay-by-use billing model. So the
not made or distributed for profit or commercial advantage and that billing model is an integral part of the system software.
copies bear this notice and the full citation on the first page. To copy 3. Hardware systems are designed at the scale of entire
otherwise, or republish, to post on servers or to redistribute to lists, containers that are managed through Virtual Machine
requires prior specific permission and/or a fee. Monitors according to the status of demand.
iiWAS2009, December 14–16, 2009, Kuala Lumpur, Malaysia.
Copyright 2009 ACM 978-1-60558-660-1/09/0012...$10.00.

512
iiWAS 2009 Short Papers Proceedings of iiWAS2009

The Cloud provides scalable resources, such as processing power processing power for different kinds of analysis, iv) worldwide
and memory, and several kinds of connectable services on a access to the processing and storage infrastructure, v) resource
transparent, virtually unlimited hardware layer [1]. This optimization, vi) be able to share the results more easily, vii) and
distributed architecture technically has many similarities with a using pricing as one more criteria for the IT infrastructure.
typical wireless sensor network as depicted in Figure 3, where a
lot of motes (see Figure 2), which are responsible for sensing and The remainder of the paper is structured as follows: section 2
local preprocessing, are interconnected with wireless connections discusses related work, section 3 presents the architectural model,
for large scale observations (see [2]). Since wireless sensor section 4 shows an example for energy monitoring, and section 5
networks are limited in their processing power, battery life and discusses advantages and drawbacks of the novel approach.
communication speed, cloud computing usually offers the Section 6 concludes the paper and summarizes the main features
opposite, which makes it attractive for long term observations and and gives a short outlook for further work.
for shared use of data and services in different kind of projects or
analysis tasks.
2. RELATED WORK
BeTelGeuse ( [3], [4]) is a platform for gathering and processing
situational data. BeTelGeuse runs on a mobile devices (e.g., a
cellular phone or a PDA), and gathers data from a body area
network over Bluetooth. It transmits the gathered data to a server
for further processing. The main advantage of BeTelGeuse is that
it can be extended to use new sensors and it runs on several
platforms. A major drawback of the BeTelGeuse is that the server
does not offer a general architecture for data processing and
Figure 2. Crossbow Mote (left) and SunSpot (right) analysis. The approach also doesn`t consider aspects like
scalability and data security.
The huge amount of data, which a sensor network is able to
deliver, demands a powerful and scalable storage and processing TinyDB [5] is a service for data aggregation in low-power,
infrastructure. Depending on the sample frequency (e.g. from 100 distributed, wireless environments. Participating motes within the
Hz or more down to few samples a day) of the sensors and the wireless sensor network are running TinyDB on top of TinyOS
defined size of a sliding data window for calculating observations [6]. It allows users to express simple, declarative queries and
the deployed infrastructure has to scale up memory, storage and have them distributed and executed efficiently in networks of low-
processing power. Today wireless sensor network platforms (e.g. power, wireless sensors. TinyDB offers a simple GUI and a Java
Crossbow MicaZ, Sentilla JCreate, SunSpot) that perform sensing API for declaring and executing queries in an acquisitional query
and complex calculations are most of the time constrained in their language, which is similar to SQL. It was used for instance on
capabilities and therefore is one appropriate way to solve this Great Duck Island off the coast of Maine, where researchers
issue offline processing of sensor data if the resources aren`t deployed a sensor network to monitor (via light, temperature, and
sufficient. infrared presence detection sensors) the nesting grounds of
seabirds [7]. 100 Crossbow Mica2 nodes stayed operational
during summer and fall of 2002, and also on summer of 2003.
TinyDB has also been used in a network with 80 motes for
microclimate analysis in the redwood trees in Santa Cruz via the
temperature, humidity, light, atmospheric pressure sensors [8].
The main advantage of TinyDB is its ease of use and efficient use
of the available processing and battery power, so that users don`t
have to be experts in wireless sensor network, because one only
has to declare the required queries. All of these deployments
collected a lot of data during several month or years of operation
and it wasn`t possible to easily share the data with other
researcher interested in the collected data.
SPINE [9] is a TinyOS-based software framework for body sensor
signal processing applications. The framework comprises signal
processing components for sensor nodes running the TinyOS [6]
environment and Java components to manage the sensor nodes
from a central coordinator node (star topology). It has further be
taken to the newly create platform DexterNet [10], an open
platform for heterogeneous body sensor networks and its
applications. The framework succeeds in abstracting from low-
level TinyOS programming, and offers a sensor data classification
Figure 3. Wireless Sensor Network service, which is operated in a distributed fashion on individual
sensor nodes and a base station computer. The approach also
The combination of wireless sensor networks, with their huge doesn`t consider aspects like scalability and data security.
amount of gathered sensor data and their limited processing
power, with a cloud computing infrastructure makes it attractive Google App Engine ( [11]) and Microsoft Windows Azure ( [12],
in terms of i) integration of sensor network platforms from [13]) are two competitors in the cloud computing business. Both
different vendors, ii) scalability of data storage, iii) scalability of provide on-demand compute and storage capabilities to host,

513
Proceedings of iiWAS2009 iiWAS 2009 Short Papers

scale, and manage Web applications and services on the Internet • Provide an interface for data export, so that the stored
hosted in a datacenter. Applications for the host environment, data can easily be taken from the Cloud storage and to
which are typically web applications, can be developed with be used in non-cloud solutions.
several different languages, ranging from Python over Java to C#.
The preferred communication method from the client`s computer • Provide a notification service, so that external
to the cloud computing environment is based on web services, services can be informed about specific events; e.g. a
which is a proven technology and supports interoperability (e.g. filter or machine learning component identifies a
between Java and .NET). The use of web services brings a lot of specific situations or has finally calculated a specific
flexibility for the development of those clients, which are hosted result.
in the customer`s environment so that the customer`s legacy • Provide data access rules, so that data, for instance
systems can be easily integrated. They can be developed in an from various customers, can be protected against
arbitrary programming language, as long web services are unauthorized use.
supported. Very important for the development of cloud
computing application is, that these applications can be developed • Provide configuration capabilities for filter chains,
locally, because both (Google App Engine and Microsoft (web) services, notifications, and data access rules.
Windows Azure) come with a local runtime environment that also
facilitates debugging before the application is deployed to the • Provide a management console for the configuration
remote cloud computing environment. of the whole system, so that it can be easily adapted
to specific needs.
3. ARCHITECTURE MODEL
For those parts of the whole systems, which are hosted at the
cloud computing infrastructure do we propose a cloud computing
model that is mainly based on pipes and filters (see [14]). The
pipes and filter design pattern (see Figure 4) is often used in
digital signal processing applications, which also fits well into the
domain of wireless sensor networks (see [15]). Pipes do not
transform data, but generally buffer data and provide a uniform
interconnection mechanism of filters. Filters do specific
processing and transformation on the input data, such as data
refinement or suppression. Filters consume data from an input and
deliver it to an output port. A simple filter could, for example be
responsible for storing the data in the database or deleting data
below or above a certain threshold. More complex filters could be
based on a machine learning components (e.g. WEKA).

Figure 4. Pipe, Filter, and Filter Chain


The general system architecture, which combines sensor networks
with the Cloud Computing paradigm, consists of several basic
services. The main requirements for the system are: Figure 5. System architecture of the proposed cloud
computing based sensor data analysis environment
• Receive and manage sensor data from a highly
scalable number of heterogeneous motes. The system architecture in Figure 5 shows the collection of base
services provided to the user as SaaS applications. The Sensor
• Manage a set or chain of filters that perform on-line Network Analytical Services group all services that are necessary
analysis on sensor data. to fulfill the requirements that where mentioned above. The
• Run filters offline on a given set of sensor data. services are based on the Cloud Provider`s infrastructure in order
to scale the demanded resources. The following base services are
• Permanently run filters on a given set of sensoric data necessary to collect and analyze sensor network data within the
in order to provide online analysis. cloud:
• Sensor Data Management: It is responsible for the
• Provide different, user definable views and management of the sensor data within the Cloud
visualizations on the sensor data and calculation Computing environment. Since the cloud computing
results. environment provide several ways for storing data, e.g.:
• Provide an interface for changing existing filters or to Google AppEngine offers Bigtables [16] for data
develop new filters out of an existing domain specific persistence and Microsoft Azure provides BLOBs,
modelling tool (e.g. KNIME). queues and tables, it’s necessary to have a flexible data

514
iiWAS 2009 Short Papers Proceedings of iiWAS2009

access layer, which raises the level of abstraction, so contains and it would provide some properties, so that the data can
that persistence mechanism can be easily exchanged. be accessed in a convenient way. Several classes could then be
• Runtime for Filter Chains: Filters are usually grouped together in a hierarchy, e.g. energy management with
configured for a specific filter chain. A filter chain is a classes like power consumption, temperature, humidity, light
reliable runtime environment for filters and it executes exposure, etc., so that polymorphism can be used in describing
various user defined filters. Such a filter chain which data has to be dispatch to the individual filter chains.
represents a specific analysis, e.g. energy efficiency
analysis of a building or a production process, where
various sensor data (e.g. temperature, gas speed,
vibrations, power/energy consumption, open/closed
windows, etc.) is continuously processed.
• Filter Chain and Filter Management: Filters and filter
chains need some management. This management
service allows the (remote) administration of filters, so
that a user is able to add new filters, delete filters and
aggregate existing filters to combined filters. Combined
Figure 6. Dispatching incoming data to specific filter chains
filters offer an aggregation mechanism for those cases,
where the number of filters gets too big or where 4. EXAMPLE: ENERGY MONITORING
specific filter/processing operations can be grouped
To show how powerful this approach can be, we show an
together. Since many filter chains are executed in example, which uses a wireless sensor network for gathering
parallel, it is necessary to offer a flexible configuration
energy consumption data of various devices and pushes this data
mechanism, which defines the following aspects: i) to the above described cloud. The basis forms a wireless sensor
which filters should be applied, ii) how is their ordering,
network (see Figure 7) built with JCreate motes from Sentilla,
iii) how big is the temporary buffer of a pipe (e.g. if a which are reading the power consumption of different devices
sliding window is applied on previous data), iv) how
from an attached 30 Amp current sensor AC&DC from Phidgets.
internal and external services (e.g. notification service) The Sentilla mote JCreate sends the sensed data in a predefined
can be accessed, and v) the dispatching of the incoming
interval (e.g. every 5 seconds) to its neighbors over the wireless
sensor data to the specific filter chains, so that the data connection link built on the Zigbee (IEEE 802.15.4) protocol. The
of several user`s isn`t mixed up and filter chains with
message being sent is called PowerMessage, and it contains the
specific analysis tasks get only the relevant sensor data. mode identification number, a message number which is
• Visualization / Views: A powerful visualization is continuously increased, and the electric potential provided by the
necessary, since the cloud computing environment current sensor. It is possible to have more data in the message, and
stores and processes a lot of data. The visualization for example to sends a number of readings in one message, which
service provides various predefined and user-defined is called in-network processing and preserves a certain amount of
views on the data and analysis results. The energy [17], but we left this out for the sake of simplicity in this
visualizations and views can be implemented with paper.
languages like OLAP and data warehouses.
• Notification Service: This service is a mechanism to
inform external applications and services about specific
situations, where it fires an event. The service itself can
provide a set of predefined and user-defined rules. Such
a rule could fire an event, if for instance, several sensor
values exceed a certain vibration threshold. This could
be for example, an indication that a gas pump failure is
approaching.
The core of the cloud computing system is built around a pipes
and filter chain approach. Since the system has to maintain many
filter chains in parallel, the reliable dispatching (see Figure 6) of
the incoming sensor data to the specific filter chains is very
important, so that the data of several user`s isn`t mixed up and
filter chains with specific analysis tasks get only the relevant
sensor data.
The dispatcher can be configured with the Filter and Filter Chain
Management. The configuration defines a mapping, in terms of i) Figure 7. Power measurement with Sentilla motes
customer/data owner so that the data of different customers can be
separated in different filter chains, and ii) sensor data type, which Since our wireless network is organized as a mesh network, each
has to be put to a specific filter chain. We assume in our individual mote forwards received messages from its neighbors
approach, that data is described and processed in an object- within its communication range of 20-50 m, depending of the
oriented manner, so that incoming data (e.g. a temperature or a actual environment. This forwarding is achieved through using the
power consumption measurement) has a data type and is derived collect-protocol provided by the Sentilla mote implementation,
from a base class. The specific class would describe the data it

515
Proceedings of iiWAS2009 iiWAS 2009 Short Papers

which eases the code to be developed for and deployed to the data can be used for further analysis and visualization. Figure 10
individual motes. shows on simple visualization of the energy/power consumption
of a customer over a given time period.
A common visualization technique is based on web applications
since cloud computing environments are operated remote to the
customer, and which frees the user from local installation. It
facilitates also the cooperation within research communities since
results can be shared more easily.

Figure 8. Sentilla USB gateway for ZigBee protocol


The mesh network contains also a Zigbee USB gateway (see
Figure 8), which is attached to a computer and collects the
messages being send from the motes. Such a ZigBee USB
gateway can also be used to deploy new application over the air
onto motes, which saves a lot of time in the setup stage, where all
motes are prepared for the energy monitoring task. Even the
communication range of an individual mote is quite limited, the
mesh networks allows us to communicated over several hundred
Figure 10. Example visualization of energy/power
meters to the ZigBee USB gateway, using the Multi-Hop RF
consumption
advantage.
In our prototypical implementation do we have a mix of languages
The computer with the attached ZigBee USB gateway manages
and platforms, which make it attractive for heterogeneous
the communication with 20 or more motes, but it could be up to
environments. The wireless sensor network platform Sentilla uses
65,000 motes on a network. The PC itself acts as a
the Java Micro Edition (JME), so that the applications deployed to
communication link between the wireless sensor network
the motes are written in Java. The wireless sensor network
observing the environment and the remote cloud computing
communication (WSNCH) hub run either on a Java virtual
infrastructure. The component in Figure 5 named wireless sensor
machine or on the .NET Common Language runtime. We
network has been refined accordingly. This component acts in our
currently have experimented with both platforms. The same
architecture as a wireless sensor network communication
situation do we have for the cloud computing environment.
(WSNCH) hub or router.
Microsoft Windows Azure prefers the .NET platform, so that we
have chosen C# for the implementation. Finally, have we used
Java as the implementation language for the Google App Engine.

5. DISCUSSION
To identify advantages and drawbacks of the proposed
combination of wireless sensor networks and cloud computing
environments we discuss it in terms of i) development issues , ii)
scalability, iii) interoperability, iv), security, v) law, vi) business
issues.
The development of such systems has the advantage, that several
Figure 9. Wireless Sensor Network Communication Hub programming languages and different kind of devices can be used
The WSNCH as depicted in Figure 9 can use for the and easily interconnected. It provides a lot of flexibility in terms
communication with the cloud computing infrastructure those of programming languages and devices, but this come with the
communication media, which are appropriate. In cases, where a drawback of heterogeneity, which quite often leads to more
PC is used and a wired internet connection exists, for example complexity.
Ethernet, the fast wired connection can be used. In cases, where The scalability of this approach seems to be unlimited, since
the wireless sensor network operates in more remote or harsh wireless sensor networks operate independently and in a mesh
environment areas, for example in a big steel or chemical plant or network, and are connected to the cloud computing environment
in urban areas, a wireless connection based on HSDPA, UMTS, or through a scalable number of wireless sensor network
even GPRS can be applied. In our case, it does not matter, if the communication hubs. The cloud computing environment itself
WSNCH is a PC/Laptop or just a Smartphone, hosting the offers a scalable infrastructure, which makes it very attractive.
relevant software components for the communication. The main bottleneck in this architecture is the remote
The configuration data, which is needed for each individual communication between the wireless sensor network
WSNCH, usually contains at least customer identification, access communication hubs and the cloud computing environment. The
authorization, location of the cloud computing infrastructure, and local computation within the cloud computing environment is
buffering settings in case of communication problems. usually fast, but the remote communication between the hub and
the cloud computing environment is much slower, which has to be
The data, which is pushed up to the cloud computing
considered in advance and further investigated in the future (see
infrastructure gets stored and processed in the previously
[18], [19]).
configured filter chains. One of the simplest filter chains would
only put the sensor data into the persistent storage. Finally, the

516
iiWAS 2009 Short Papers Proceedings of iiWAS2009

The interoperability between different kinds of devices, services, [2] Feng Zhao and Leonidas Guibas, Wireless Sensor Networks -
and programming platforms is organized on the level of web An Information Processing Approach.: Morgan Kaufmann,
services interfaces, which provides a good level of abstraction and 2004.
is a proven concept. But it must be noted, that the necessary
interoperability measures usually lead to some overhead, which [3] Petteri Nurmi, Joonas Kukkonen, and Eemil Lagerspetz,
may have an impact on the overall system performance (see [18]). "BeTelGeuse – A Tool for Bluetooth Data Gathering," in
Proceedings of the ICST 2nd international conference on
Several measures are taken to ensure security of the sensed, Body area networks , Florence, Italy , 2007.
processed and stored data. The data store and processed in cloud
computing infrastructure are protected by authentication and [4] Joonas Kukkonen, Eemil Lagerspetz, Petteri Nurmi, and
authorization measures, and can if needed be encrypted. But Mikael Andersson, "BeTelGeuse: A Platform for Gathering
Foster [20] mentioned, that currently, the security model for and Processing Situational Data," IEEE Pervasive
Clouds seems to be relatively simpler and less secure than the Computing, vol. 8, no. 2, pp. 49--56, Aug. 2009.
security model adopted by Grids. The sensor data transferred from
the wireless sensor network communication hub to the cloud [5] Samuel Madden, J. Franklin, Joseph M Hellerstein, and Wei
computing infrastructure can be secured by TLS/SSL, which is a Hong, "TinyDB: An Acqusitional Query Processing System
proven technology. Last but not least can the data communication for Sensor Networks," in ACM Transactions on Database
within the wireless sensor network also be secured with a 128-bit Systems, 2005, p. 47.
AES encryption. Since motes in a wireless sensor network have [6] Philip Levis et al., "TinyOS: An Operating System for
limited energy resources (i.e. are battery powered), and they Wireless Sensor Networks.," in Ambient Intelligence., 2005.
sometimes don`t process in some cases mission critical data, it`s
possible to omit the encryption overhead if necessary. [7] R. Szewczyk, A. Mainwaring, J. Polastre, and D. Culler, "An
analysis of a large scale habitat monitoring application," in
The storage of data outside the corporate environment or outside Proceedings of the Second ACM Conference on Embedded
the state/country is sometimes strictly regulated in company Networked Sensor Systems (SenSys), 2004.
policies and laws. Some institutions like the Michigan State
University clearly state in a document (see [21]) defining the use [8] G. Tolle et al., "A macroscope in the redwoods," in
of cloud computing services as follows: “Cloud-computing Proceedings of the Third ACM Conference on Embedded
services must NOT be used with any of these confidential data Networked Sensor Systems (SenSys), 2005.
types, unless an appropriate contractual agreement can be
negotiated with the service provider by the University”. Such [9] R. Giannantonio, F. Bellifemine, and M. Sgroi, "SPINE -
policies sometimes prevent the adoption of remote cloud (signal processing in node environment): A framework
computing services in specific projects. forhealthcare monitoring applications based on body sensor
networks," 2008.
Beside of the technical issues offers the combination of cloud
computing and wireless sensor network a lot of opportunities for [10] Philip Kuryloski et al., "DexterNet: An open platform for
new business models (see [1]). But technical risks of wireless heterogeneous body sensor networks and its applications," in
sensor networks and cloud computing must not be Proceedings of 6th International Workshop on Body Sensor
underestimated. Networks, Berkeley, USA, 2009.

6. CONCLUSION [11] Charles Severance, Using Google App Engine , O'Reilly, Ed.:
We have presented a novel way of combining wireless sensor O'Reilly, 2009.
networks with cloud computing services. The wireless sensor [12] Joshy Joseph, "Cloud Computing: Computing: Patterns For
networks are used to sense and collect environmental data. Since High Availability, Scalability, And Computing Power With
wireless sensor networks are limited in their processing power, Windows Azure ," MSDN Magazine, May 2009.
battery life and communication speed, cloud computing usually
offers the necessary storage capacity and processing power for [13] Elisa Flasko, "Data Services: Access Your Data On Premise
long term observations, analysis and use in different kind of Or In The Cloud With ADO.NET Data Services," MSDN
environments and projects, since the basic infrastructure remains Magazine, May 2009.
the same. The presented concept and implementation must now be
further evaluated in different use cases to measure its advantages [14] Erich Gamma, Richard Helm, and Ralph E. Johnson, Design
and shortcomings. Further work should investigate the integration Patterns. Elements of Reusable Object-Oriented Software.:
of predefined complex filters and different classifiers and define a Addison-Wesley Longman, 1995.
suitable ontology for semantic reasoning, so that data analysis can [15] Werner Kurschl, Stefan Mitsch, and Johannes Schönböck,
be done within the cloud computing environment. "Modeling Distributed Signal Processing Applications," in
Proceedings of 6th International Workshop on Body Sensor
7. REFERENCES Networks, Berkeley, USA, 2009.
[1] M. Armbrust, A. Fox, R. Griffith, A. Joseph, R. Katz, A.
Konwinski, G. Lee, D. Patterson, A. Rabkin, I. Stoica, and [16] Fay Chang et al., "Bigtable: A Distributed Storage System
M. Zaharia,., "Above the Clouds: A Berkeley View of Cloud for Structured Data," in OSDI06: Seventh Symposium on
Computing," University of California, Berkeley, Operating System Design and Implementation, Seattle, WA,
UCB/EECS-2009-28, 2009. 2006.

517
Proceedings of iiWAS2009 iiWAS 2009 Short Papers

[17] K. Römer, "Time Synchronization and Localization in 1-10, 2008.


Sensor Networks," Swiss Federal Institute of Technology
Zürich (ETH Zürich), Zürich, Schweiz, Phd Thesis 2005. [21] David Gift, "Appropriate Use of “Cloud Computing”
Services by the Michigan State University Community,"
[18] F. Douglas, A. Mirno C. Loosley, High-Performance Michigan State University, WORKING DRAFT 2008.
Client/Server.: John Wiley and Sons Inc., 1997.
[22] C. Hewitt, "ORGs for Scalable, Robust, Privacy-Friendly
[19] Peter J. Denning, "The Locality Principle," Client Cloud Computing," IEEE Internet Computing, vol. 12,
COMMUNICATIONS OF THE ACM, vol. 48, no. 7, pp. 19- pp. 96-99, 2008.
24, July 2005.
[20] Ian Foster, Yong Zhao, Ioan Raicu, and Shiyong Lu, "Cloud
Computing and Grid Computing 360-Degree Compared,"
Proc. Grid Computing Environments Workshop GCE '08, pp.

518

S-ar putea să vă placă și