S p t
t
serial
parallel
( )
=
(2)
where,
t
parallel
is the entire time needed to finish the full pro-
cedure by a parallel algorithm using
p
processors, and
t
serial
is
the corresponding time required by serial processing. In these
experiments, serial processing only using one processor is a
controlled case where block partition and the recursive pro-
cedure are embodied, but parallel mechanism is exempted.
The single processor executes the recursive procedure (as
previously described) to generate final results for each block,
and then writes the results to the output file instead of send-
ing them to another processor, block-by-block. The setting of
block partition and the recursive procedure in serial process-
ing are the same as parallel processing. The execution times
used in Equation 2 represent the entire time required to finish
the full procedure of an algorithm, including file reading,
writing time, and computing time.
The experiments using different numbers of processors,
which can be specified by user but cannot be more than the
total number of cores (or virtual cores) available in the em-
ployed computer, are carried out. For both computers shown
in Table 2 and Table 3, even numbers under the total number
are used, which are as indicated in Table 4 and Table 5. The
Plate 2. Fusion results by block-regression-based algorithm. Fusion results and the multispectral images are displayed in the form of
false infrared (
ir
) color composites of 4, 3, and 2 bands: (a) and (b) Two segments of the QuickBird panchromatic image; (c) and (d) Two
segments of the QuickBird multispectral image; and (e) and (f) Two segments of the fusion result of the BR algorithm.
T
able
2. C
onfigurations
of
the
W
orkstation
C
omputer with
W
indows
-7 OS
Dell Precision T7500 workstation
CPU: Two Intel Xeon X5675 CPUs, 3.06 GHz, 6 cores per CPU,
24 virtual CPU cores using hyper-threading technology
RAM: 48 GB
Disk and file system: A Solid-State Drive (SSD) with 120 GB,
NTFS file system
OS: Windows-7 Professional, Service Pack 1, 64-bit
Compiler: Microsoft Visual C++ 10.0
MPI: Intel MPI 4.0 Update 3, IA32
T
able
3. C
onfigurations
of
the
W
orkstation
C
omputer with
L
inux
OS
A customized workstation
CPU: Two Intel Xeon E5520 CPU, 2.26 GHz, 4 cores per CPU, 16
virtual CPU cores using hyper-threading technology
RAM: 12 GB
Disk and file system: Areca 1222 RAID card, raid 0, 6 disks, Strip
Size: 4 KB, SAS Disk: Seagate Savvio (ST9300603SS), 300G, 10K,
ext3 file system
OS: Redhat Enterprise Linux Server, Release 5.4, X86_64
Compiler: GNU GCC 4.1.2
MPI: LAM-MPI 7.1.1
380
May 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING