ABCI
Home Page
■
I benchmarked ABCI_MP on the following three PCs with different numbers of
CPU cores by running a sample file called LEP.abc:
Intel
Core 2 Duo E6600 @2.4GHz: Windows XP: 2GB RAM CPU
time used=21.7s
Intel
Core 2 Quad Q6600 @2.4GHz: Windows XP: 4GB RAM CPU
time used =13.3s
Intel
Core i7 920 @2.67GHz: 64-bit Windows Vista: 3GB RAM CPU time used =5.1s (ABCI_MP 32-bit version)
Intel
Core i7 920 @2.67GHz: 64-bit Windows Vista: 3GB RAM CPU time used =4.6s (ABCI_MP 64-bit version)
Findings:
1.
Performance of ABCI_MP scales almost linearly with the number of CPU cores.
2.
ABCI_MP 64-bit version runs 10% faster than the 32-bit version on 64-bit
Windows Vista.
Running
ABCI_MP 64-bit version on Windows Vista 64-bit and Corei7 looks very promising.
■ Now, GUI for ABCI to PyPi is available and ABCI can
be easily installed on GNU/Linux and Windows,
thanks to Dr. Sergey V. Matsievskiy.
Installation guide is available here.
■
Now, the Linux versions of ABCI_MP are available, thanks to Drs. Yong-Chul Chae and Xiaowei
Dong of
ANL
and Dr. Jonathan Smith of
visit
the installation guide page:
■ Now, the Windows XP 64 bit version of ABCI_MP
is also available. It is here:
http://abci.kek.jp/ABCI_MP64_12_5.zip
Dear
ABCI users,
ABCI_MP
was updated to version 12.5. This version fixed some small bugs for users who
want to
use
a large amount of meshes. If you do not belong to this category, you may keep
using the version 12.3,
which
is also included in the package.
It is here:
http://abci.kek.jp/ABCI_MP_12_5.zip
ABCI_MP
was updated to ABCI_MP_12.3 (version 12.3). The new or improved features
include
the transverse extension of Napoly integral (derived by Shobuda)
so that ABCI can now
handle
calculations of transverse wake potentials in
structures having unequal tube radii at the two sides,
still keeping the integration path confined to a
finite length by having the integration contour beginning
and ending on the beam tubes. More details are
described in the paper THPAN036
presented at PAC07.
Improvement of the open boundary condition.
ABCI used to adopt
the conventional open
boundary condition where all waves propagating in the
beam pipe are assumed to have the phase
velocity equal to the speed of light. But in general
cases, the propagating fields can be represented
as a linear superposition of the waveguide modes and
each mode has its own phase velocity which
varies in frequency. Aharonian
et al. introduced a more advanced formula for the open boundary
conditions in the DBCI code and ABCI now adopts it. In
this method, the phase velocities of
all the travelling waveguide modes are represented
correctly in the code.
ABCI_MP
supports parallel processing in OpenMP for shared-memory computers,
namely
a PC with several CPUs (e.g., 8 AMD Opterons) or a
CPU with multiple cores
(e.g.,
Intel Core2Duo), which share the same memory. It also supports multi-threaded
shared-memory
system. Tests with a Core2Duo PC (two cores) show that ABCI_MP is
about
1.7 times faster than a non-parallelized ABCI. ABCI_MP also adopts the dynamic
memory
allocation for nearly all arrays for field calculations so that the amount of
memory
needed
for a run is determined dynamically during runtime. You can use any number of
meshes
as
far as the total allocated memory is within a physical memory of your PC (if it
exceeds,
the
ABCI_MP starts to access a hard disk and the computation will be slowed down
severely,
although
it still runs). The size of physical memory needed to run a job with N million
meshes is
approximately
0.1*
DOS
window so that you can see the progress of a job on screen.
The
MPI version of ABCI is
also under consideration.
Inside,
there are four folders and their names tell you what they are.
You
can place the application programs anywhere.
You
can also place the input file folder anywhere, not necessary in the same folder
as
ABCI executable modules. No
installation of the program is necessary.
The
recommended use may be as follows:
1.
Place the ABCI
application folder in the Program Files folder.
2.
Then, create the short-cut(s) of the ABCI executable module(s) on the desktop.
3.
Place the input file folder anywhere.
4.
To run ABCI,
you just drag and drop the input file (such as sample1.abc) on top
of
the ABCI short-cut, and then the
DOS window appears, saying that ABCI is now running.
5.
When the computation ends, you will find the output files in the same folder as
the
input file. Very simple, is it not?
You
should better assign Notepad or Wordpad programs to
open these files.
The
alternative way is to double click one of the ABCI applications, and the DOS window
appears,
asking the name of the input file. If the input file is located in the same
folder
as
the ABCI executable module, you
just type its name such as sample1.abc. If not,
you
have to give the full path-name (for example, C:\Input_files\sample1.abc).
The
recommended way looks more convenient to me.
I
enclose the TopDrawer program for Windows. You just
drag and drop a
TopDrawer file on it and then all figures pop up. If you want
to convert them to
postscript
files, just right click the figure and choose PostScript option.
You
should better install Ghostscript and Ghostview programs to view and print
postscript
files from here:
http://www.cs.wisc.edu/~ghost/index.htm
If
you need to compile the ABCI
source code using the Compaq Visual Fortran,
you
need to add /fpscomp:filesfromcmd option in the
compiler so that the command-line DOS
window
appear to ask you the input file name at run time. The Windows version of ABCI codes are
the
stand-alone versions, and there is no need to link with other subroutines.
The
Windows version of ABCI is
very fast, and together with TopDrawer for Windows,
you
can do all necessary jobs only on Windows.
This
information and the ABCI
programs are free to use, but it is not Open Source. If you find someone
who
want to use it, just tell him the above URL to download the package.
The
animation of the electric fields in the KEK ARES cavity is available now to
download.
But,
watch out that the file is big (96MB):
Dr.
Iker Rodriguez has kindly provided us two compressed files for the ARES cavity
animation:
Intel
Indeo codec 5.10 version (1.7MB):
http://abci.kek.jp/aresindeo.avi
DiVX version (2.6MB):
http://abci.kek.jp/aresDivX.avi
Please
give me your comments.
Best
wishes,
Yong
Ho Chin
yongho.chin@kek.jp
All questions
regarding to this home page should be addressed to
Yong Ho Chin (yongho.chin@kek.jp)
Last updated on May 13, 2009 by Y. H. Chin