problem, the creation of the maps has been based on a HPC pilot
application running on Microsoft Windows HPC server. HPC provides
the computing power that is required for the fire ignition and weather
map calculations because of the large geographical extent and
the high spatiotemporal resolution. A computation cluster is used,
composed of two quad-core processing nodes (one head node and
one computing node). The map resolution for both sequential and
parallel execution was the same, i.e. 500 m. For the fire ignition probability
maps the “task flow” model was used where a set of unlike
tasks are run in a prescribed order, usually because one task depends
on the result of another task. Whenever, two or more tasks can run
independently then they are assigned to a separate core and run concurrently.
By following the HPC approach, the computation speed was
increased of about 33%, by activating up to five cores simultaneously.
More specifically, processing mean time of sequential processing was
initially 333 s, while HPC decreased the time to 224 s. Even if these
durations are considered relatively small, the benefit of HPC would
emerge for higher resolutions and greater geographic areas, where
the processing times would be much higher.
Regarding the forecasted weather mapping, a total of 85 raster
weather maps and 32 shapefiles are created by using “parametric
sweep” implementation and initializing all 8 available cores. The “parametric
sweep” model consists of multiple instances of the same application
that run concurrently, and the input and output are a set of indexed
files. The serial execution takes approximately 14 min to be completed,
while 12 min is required for the main processing and the rest is the preand
post-processing of the data. HPC reduced the execution time to
5 min for the whole process and 3 min for the main process, i.e. 64%
and 75% speed increase, respectively. One of the disadvantages of the
HPC is the limited efficiency under usual network (i.e. fast Ethernet)
configuration. The four cores of the computing node took more than
double the time to complete weather mapping, compared to the head
node, probably due to the communication performance (e.g. latency,
overhead and bandwidth) (Martin et al., 1997)
problem, the creation of the maps has been based on a HPC pilotapplication running on Microsoft Windows HPC server. HPC providesthe computing power that is required for the fire ignition and weathermap calculations because of the large geographical extent andthe high spatiotemporal resolution. A computation cluster is used,composed of two quad-core processing nodes (one head node andone computing node). The map resolution for both sequential andparallel execution was the same, i.e. 500 m. For the fire ignition probabilitymaps the “task flow” model was used where a set of unliketasks are run in a prescribed order, usually because one task dependson the result of another task. Whenever, two or more tasks can runindependently then they are assigned to a separate core and run concurrently.By following the HPC approach, the computation speed wasincreased of about 33%, by activating up to five cores simultaneously.More specifically, processing mean time of sequential processing wasinitially 333 s, while HPC decreased the time to 224 s. Even if thesedurations are considered relatively small, the benefit of HPC wouldemerge for higher resolutions and greater geographic areas, wherethe processing times would be much higher.Regarding the forecasted weather mapping, a total of 85 rasterweather maps and 32 shapefiles are created by using “parametricsweep” implementation and initializing all 8 available cores. The “parametricsweep” model consists of multiple instances of the same application
that run concurrently, and the input and output are a set of indexed
files. The serial execution takes approximately 14 min to be completed,
while 12 min is required for the main processing and the rest is the preand
post-processing of the data. HPC reduced the execution time to
5 min for the whole process and 3 min for the main process, i.e. 64%
and 75% speed increase, respectively. One of the disadvantages of the
HPC is the limited efficiency under usual network (i.e. fast Ethernet)
configuration. The four cores of the computing node took more than
double the time to complete weather mapping, compared to the head
node, probably due to the communication performance (e.g. latency,
overhead and bandwidth) (Martin et al., 1997)
การแปล กรุณารอสักครู่..
![](//thimg.ilovetranslation.com/pic/loading_3.gif?v=b9814dd30c1d7c59_8619)