The document discusses grid computing in remote sensing data processing. It describes how grid computing can help process huge amounts of remote sensing data in real-time by distributing processing across networked computers. Key requirements for a grid environment for remote sensing include sharing computational and software resources, managing resources, and supporting different data formats. Case studies demonstrate how grid middleware can improve the efficiency of tasks like image deblurring and generating lookup tables for aerosol analysis.
2. Contents
Introduction to Grid computing
Requirements for remote sensing data processing on grid environment
Advantages for remote sensing systems to employ a grid computing
The structure and character of system based on grid
Grid middleware for remote sensing image processing
Literature review
Case study1
Case study2
Summery
References
3. Introduction
A huge amount of remotely sensed data is acquired every day.
These Remote sensing images need to be processed and stored in real
time or near-real time.
Processing the huge amount of data with complicated algorithms
needs great computational power beyond a single machine because it
is not only data intensive, but also computation intensive.
Grid computing is a promising technology that seeks to easily and
efficiently coordinates the sharing of geographically distributed
computing resources, thereby bring supercomputing power to its
users.
Some grid projects have been carried out, such as Globus, Condor
and Unicore.
4. Requirements for remote sensing data processing on grid
environment
Main objective is to design and build a Grid-Based Problem Solving
Environment to allow sharing of software and computational
resources for remote sensing data processing among different
organizations, through high speed networks.
The following requirements are identified for this grid environment:
Sharing of computational resources
Sharing of software resources
Management of resources.
Secure usage of the services available on the grid in agreement with
the access policies.
Efficient usage of the network links.
Composing and compiling of new processing applications upon the
cooperation of existing ones.
Supporting different data formats.
5. ATYPICAL VIEW OF GRID
ENVIRONMENT
User
Resource Broker
Grid Resources
Grid Information Service
A User sends computation
or data intensive application
to Global Grids in order to
speed up the execution of
the application.
A Resource Broker distribute the
jobs in an application to the Grid
resources based on user’s QoS
requirements and details of available
Grid resources for further executions.
Grid Resources
(Cluster, PC, Supercomputer, data
base, instruments, etc.) in the
Global Grid execute the user jobs.
Grid Information Service
system collects the details of
the available Grid resources
and passes the information
to the resource broker.
Computation result
Grid application
Computational jobs
Details of Grid resources
Processed jobs
6. Advantages for remote sensing systems to employ a grid
computing
• Cost savings through the sharing of resources
• Scalability to meet variations in resource demands and balance
work loads
• Shorter time to results
• Enable collaboration across organizations and among widely
distributed users, by sharing resources (data, software, and
hardware).
• More efficient use of available resources
• Increased productivity through standardization and access to
additional resources.
7. The Structure of Remote sensing processingsystem based on Grid
The overall architecture of remote sensing processing system based
on Grid and service is composed of a structure which includes the
following four layers
Data layer it can offer various raw products to processing service
unified system upon users’ request for
further and deeper processing results.
Processing layer is formed by
various services for remote sensing
processing whose package mode are
Grid services based on high-performance
environment .
Figure shows the layered system
architecture and its relationship
8. Grid layer, the main role it plays is
managing both the data and service resources so that to achieve the
goal of sharing the resources,
providing a unified system operating environment of remote
sensing application and coordinating the tasks of remote sensing
processing.
Application support layer.
Its role is the connectivity between the remote sensing systems
facing end-users and various sources of remote sensing processing
in Grid.
It is formed by middleware supporting Grid, Tools such as
automatic generation tools of remote sensing application, user
interfaces, application deployment tools, etc.
9. Grid Middleware for remote sensing image Processing
Grid middleware for remote sensing image processing to meet the
needs of real time or near-real time processing.
The middleware is an advanced high throughput Condor-based
computing grid middleware that supports
The execution of remote sensing image
Processing over a geographically distributed, and
Heterogeneous collection of resources.
Condor system provides the technologies for resource discovery,
dynamic task-resource matching, communication, and result
collection, etc.
11. Functions of each sub-modules:
Tasks partition --user can input the number of tasks partitioned, and
then the tasks will be auto-partitioned.
Task submission --user can submit all the partitioned tasks to the grid
manager.
Task execution --include remote sensing image processing programs,
when the tasks are submitted, relative programs will be remotely
called by the grid nodes for the task execution.
Task monitoring --used to monitor the image processing progress,
Result collection --all the execution results on each grid node will be
auto-collected,
Result piecing --all the returned results will be merged into one final
whole image, which is the final result of image processing just like
that obtained on a single computer.
12. Literature review
• Chaolin Wu et.al.(2005) analysed a way to solving the large-scale or complex
problems through grid computing.
• To apply the Grid computing to remote sensing information, a Remote Sensing
Information Grid analysis and Service Node (RSIGN) has been established in
Tele-geoprocessing Group, Institute of Remote sensing Applications, Chinese
Academy of Sciences.
• The management and distribution of the tasks are key problems in RSIGN node.
How to distribute and manage the tasks has significant influence on the efficiency
of the whole Grid system.
• Also discussed on two main strategies for RSIGN node: geometric parallel and
algorithm parallel. and
• How to choose the best task managing strategy for different problems in remote
sensing information processing and analysis.
13. Yong Xue, Tong Yu et al.(2006) proposed,
A New Approach to Generate the Look-Up Table for Aerosol Remote Sensing on
Grid Platform.
LUT is popular used in aerosol remote sensing retrieval which is a matrix of
dependency variable corresponding a set of combinations of independent variable
values.
They used 6S module to generate LUT. And Also discuss approaches to
parameterization, task partitioning, generated methodology, and the collection of
result.
Jing Dong et.al. (2012) designed
The cluster as a hierarchical one, and achieved a Global Task Scheduler for the
workflow process, improved the system performance.
It is One of the most effective solutions is Grid Computing platform built on the
basis of network, which is a scalable virtual unified platform with unlimited
computing power and storage capacity.
The RSSN (Remote Sensing Information Service Grid Node) is a PC cluster for
Remote Sensing Quantitative Retrieval. It was used for producing aerosol optical
depth (AOD) production covered the land area of Asia.
The Computing pool was built with common commercial PCs instead of Servers
or Workstations because of their price advantage. The PCs were connected as a
loosely coupled LAN via Network Switches.
14. Casestudy1
Analysis of image de-blurringdistributedprocessing
• To test the performance of distributed image deblurring on a
grid, the first experiment was conducted on a grid testbed of 4
commodity PCs nodes connected using 100Mbps Ethernet
switch have the same performance.
• The image data used for their experiments is 128Mb satellite
remotely sensed image.
• The geometric parallel strategy is taken in our distributed
processing experiments.
15. • Figure shows The overlap strategy of image division. (Grey area is the
neighbourhood pixels for sub-image1)
• The whole image was partitioned into 4 sub-images as equally as possible,
which is same to the number of grid nodes.
• To check the performance of our grid middleware and grid computing test
bed, a comparison is made between our work and a traditional computational
test on a single computer conducted with only one grid node of the test bed.
16. Fig shows The comparison between the
original image (upper) and de-blurred
image (bottom) on grid test bed
Table shows the performance for 4 tasks
partitioned and 4 available grid nodes of
grid computing testbed-1 in their
preliminary experiment.
17. • The execution time on testbed-1 Tg is much less than that on a single
computer Ts, which demonstrates the good performance of thier grid
middleware for improving the efficiency of image deblurring
processing.
• Conclusion:
• By means of image processing grid middleware, grid nodes can be
aggregated to meet the needs of intensive remote sensing image
processing computations. Using grid computing platform, real time or
near real time remote sensing image processing may be realized
18. Casestudy2
ANewApproachto Generate the Look-UpTable forAerosolRemote Sensing
on Grid Platform
• This case study focuses on realization of the compute-intensive look-
up table generation on GCP-ARS (Grid Computation Platform for
Aerosol Remote Sensing), which was one grid middleware that are
developing based on Condor system.
• GCP-ARS architecture mainly consists of three entities: clients, a
resource broke, and producers.
• Aerosol retrieval is launched though a client GUI for execution at
producers that share its idle cycles through the resource broker.
• The resource broke manages tasks application execution, resource
management and result collection by means of Condor system.
19. Methodology
For aerosol retrieval, they refer to apparent reflectance matrix
corresponding to Sensor-Target-Solar geometric conditions, ground
surface reflectance, aerosol and atmosphere conditions.
6S methodology is selected for LUT generation, which is Radiative
Transfer Models (RTM) for atmospheric radiative calculation.
6S codes can simulate the radiance and reflectance in the solar
spectral region (0.25- 4.0um) considering the scattering and absorbing
atmospheric effects due to gases and aerosol, as well as the surface
effects.
• The generation of LUT is actually time-consumed.
• Building one LUT of 91125 input combinations sets needs about 2
days on one PC.
• Figure 4.2.1 show the flow chart to generate the LUT.
20. Parameterization include
Radiative Transfer Model
parameters input, LUT structure
parameters such as maximum,
minimum value and of each
independent variable, which
decide the structure and size of
the LUT, tasks partition
parameters, execution program
(default is 6S codes) selection.
Execution phase refer to the
computation of dependency
variable by execution program.
The final LUT will be saved at
termination
21. EXPERIMENTSANDANALYSIS
Experiments were carried out on a grid computing pool consists of 5 low
commodity PCs nodes connected using 100Mbps Ethernet switch.
Their test is to generate one Apparent Reflectance LUT corresponding to
independent variables: solar zenith, sensor zenith, relative zenith, ground
surface reflectance and aerosol optical thickness (AOT).
Table shows the structure information of LUT generated
22. Table shows the performance for varying number of tasks partitioned
and producers.
The sequential execution time(Ts) was from test on a single computer.
The execution time (Tn) decreases with the number of tasks
partitioned increasing when the number of available producers is more
than number of tasks partitioned due to more producers contribute to
the execution at a time.
23. CONCLUSION:
• LUT could be quickly set up through grid computing
technique, which, by pooling and aggregating the idle CPU circle
together, has high throughput computing to meet the needs of
intensive computations.
• The LUT generated by 6S module could be used for GCP-ARS to
retrieve the aerosol optical thickness (AOT).
24. Summery
• Grid platform for Distributed Remote sensing Image processing are
very encouraging, which show the good potential of grid computing
technology for satellite image processing.
• By means of image processing grid middleware, grid nodes can be
aggregated to meet the needs of intensive remote sensing image
processing computations.
• Using grid computing platform, real time or near real time remote
sensing image processing may be realized.
25. REFERENCES
• Giovanni Aloisio, Massimo Cafaro, Italo Epicoco, Gianvito Quarta; “A
Problem Solving Environment for Remote Sensing Data Processing”,
Proceedings of the International Conference on Information Technology:
Coding and Computing (2004)
• Jiakui Tang, Aijun Zhang, Shengyang Li “Preliminary Research on Grid-
based Remote Sensing Image distributed Processing” IFIP International
Conference on Network and Parallel Computing – Workshops,2007.
• L. Fusco, R. Cossu, and C. Retscher. “Open Grid Services for Envisat and
Earth observation applications,” in High Performance Computing in Remote
Sensing, A. Plaza and C. Chang (Eds.), Chapman & Hall, Taylor & Francis
Group, 2008
• Liang zhonga & hongchao maa ” research of fast processing and
distribution remote sensing image based on the grid technique” the
international archives of the photogrammetry, remote sensing and spatial
information sciences. vol. xxxvii. part b4. beijing 2008.
26. • Tang, J.K., Xue, Y., Guan, Y.N.et.al.”A New Approach to Generate
the Look-up Table Using Grid Computing Platform for Aerosol
Remote Sensing”, 2004 IEEE International Geoscience and Remote
Sensing Symposium.
• Tianhe Yinet.al. “Eco-environmental Information Grid Platform and
Its Implement” Ministry of Science & Technology of Zhejiang, 2010.
• Wan wei1, yong xue “A high performance remote sensing retrieval
application on an institutional desktop grid” China Center for
Resource Satellite Data and Application, Beijing 100830, China. 2009
In fact, the amount of data to be transferred from end users to nodes and vice-versa is quite large. The data movement, considering that it can be a bottle neck, impacts on the global performance of the architecture.