The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
472
Figure 4 and Figure 5 depicts the time spent for distributed
orthorectification using several numbers of nodes.
Distributed Orthorectification in LOA
Number of node{s)
bob Stand-alone in LPS
rza Distributed Computing
Trend
у = 0 4278x-° ““
R : = 0 9978
Figure 4. Orthorectification time in hours and minutes for 240
images
Distributed Orthorectification in LOA
Number of node(s)
a Distributed Computing
Trend
y = 36 019x' 3 9525
R 2 = 0.9976
Figure 5: Orthorectification time in hours and minutes for 858
images.
For Block 240, orthorectification in stand-alone mode using
LPS was performed on one of the ortho nodes and took 10
hours 20 minutes. The source data is located on the file server
and for stand-alone mode the ortho images created were written
to the local hard drive. Orthorectification in the distributed
computing environment using 1 node took 10 hours 09 minutes.
Using 5 nodes accelerated the orthorectification time by factor
4.2 to 2 hours 27 minutes.
For Block 858, orthorectification using 1 node took 1 day 12
hours 37 minutes. Using 5 nodes decreased the processing
time by a factor of 4.6 to 7 hours 59 minutes.
5. RESULTS SUMMARY
This experiment showed that distributed processing can be very
beneficial when working with large orthorectification projects.
As could be seen in Figure 4 and Figure 5, processing time
decreases significantly with the number of nodes used. It is
clear that distributed processing will be more efficient with
larger ortho projects.
Orthorectification is accelerated by a factor of 4.2 to 4.6 times
using a 5 node cluster environment.
When file caching is turned off for the 5 node run, it slowed
down the process by only 3%. File caching might be necessary
in order to prevent file writing errors.
The system setup used for this experiment is sufficient.
However it is not the ideal setup because the experiment was
not done in a dedicated network. The ideal setup would be all
ortho nodes; server and client machines and the data storage are
in a dedicated network. This will prevent the network traffic
that has been examined during this experiment. Also, the results
would be faster and more reliable. However bottle-necks can
always happen even in a dedicated network.
As a future studies, this experiment will be repeated using a
dedicated network.
REFERENCES
Graham, L. 2006. Enterprise Geospatial Production Revisited,
Photogrammetric engineering and remote sensing, vol. 72, n°5,
pp. 486-492
ERDAS Ortho Accelerator,
http://www.erdas.com/products/Product_Rdr.aspx7CURRENTI
D=114
GeoCue Corporation, http://www.geocue.com/