Nigel Goddard & Greg Hood
Pittsburgh Supercomputing Center
pgenesis@psc.edu
(Updated March 16, 2006 for PGENESIS 2.3)
(Updated March 21, 2003 for PGENESIS 2.2.1)
NOTE: Before installing this software, please check that you have the latest version by checking the GENESIS web or ftp site:
http://www.genesis-sim.org/GENESIS/ or ftp://genesis-sim.org/pub/genesis/ We *strongly* advise you to download the latest version if you don't already have it. Additional information about PGENESIS may be obtained from the PGENESIS home page:
http://www.psc.edu/Packages/PGENESIS
The PGENESIS e-mail address is pgenesis@psc.edu.
(The serial GENESIS email address is genesis@genesis-sim.org, the GENESIS web page is located at http://www.genesis-sim.org/GENESIS, and the ftp site is ftp://genesis-sim.org/pub/genesis/.)
PGENESIS (Parallel GENESIS) is designed to allow researchers to run simulations faster on parallel hardware; it supports both running simulations of large networks on multiple processors, and running many simulations concurrently (e.g. for parameter searching). PGENESIS will also run on a single workstation and this can be useful for testing.
If you have a network model and want to use PGENESIS, you will have to partition your model so that specific neurons are created on each of the PGENESIS processes, and then connect them up using the PGENESIS raddmsg or rvolumeconnect script functions. Unfortunately, with PGENESIS 2.3 this is not automatic, although we hope to have resolved this issue in a future version of GENESIS. Large single cell models cannot be split across processes in PGENESIS 2.3, though, again, this is an issue we are looking at for future versions of GENESIS.
PGENESIS will run on almost any parallel cluster, SMP, supercomputer, or network of workstations where MPI and/or PVM is supported, and on which serial GENESIS itself is runnable. It is much easier to run on a network of workstations if they have a shared file system so that any workstation can access any file. If this is not the case, you will need to install PGENESIS independently on each workstation and you will need to have a copy of the files required by your simulation (data files, script files, etc) on each workstation, preferably in an identically named directory path.
PGENESIS adds a library (parlib) to an installed version of GENESIS, creating an executable "pgenesis" that has the usual GENESIS objects and commands, plus a "postmaster" object (post) and some other new objects and commands. The "postmaster" objects communicate with each other through a standard interprocess communication library such as MPI or PVM.
For more complete information on how to use PGENESIS, see the PGENESIS webpage: http://www.psc.edu/Packages/PGENESIS.
cat genesis-2.3-src.tar.gz | gunzip | tar -xvf -
This will produce a directory genesis-2.3/genesis. Follow the instructions in genesis-2.3/genesis/src/README. Make sure that after doing a "make" or "make nxall" you also do a "make install" or "make nxinstall".
echo version | pvm
Make sure that your environment includes the proper settings for $PVM_ROOT
and $PVM_ARCH, and that the PVM object libraries are in
$PVM_ROOT/lib/$PVM_ARCH.
If you want to install PVM, you can get the latest version, as well as
more information about PVM from the PVM home page:
http://www.epm.ornl.gov/pvm/pvm_home.html
You do not need the root password to install PVM. After untarring the distribution, you will find installation instructions in pvm3/Readme. A PVM manual is is also available from the PVM home page. A PVM distribution and manual is also available from the GENESIS ftp site, but it is not guaranteed to be the most recent version. See the PVM document Readme.mp for more specific issues about running PVM on multiprocessor machines.
'make nxinstall' - to compile and install pgenesis without Xodus
'make install' - to compile and install pgenesis with Xodus
'make extended' - to install extended pgenesis with Xodus
'make nxextended' - to install extended pgenesis without Xodus
'make env' - to show some of the configuration settings
'make clean' - to remove this machine type's object files
'make spotless' - to remove all machine types' object files
Normally, you will use the 'make install' command, which compiles and installs PGENESIS with Xodus. Once the Makefile has been customized, execute:
make install
If you do not use Xodus, first make nxgenesis for serial GENESIS, and
then execute "make nxinstall".
If there are errors that are not ignored, then redo this command piping the output into a file and submit it as a bug report to genesis@genesis-sim.org:
make install > & ! make.out &
(or for the Bourne shell: make install > make.out 2>&1 & )
Note that, unlike the case with serial GENESIS, "make install" or "make nxinstall" is used to both compile and install PGENESIS.
The compile process can be viewed by typing:
tail -f make.out [ hit ^C to exit the tail ]
Note that this is done in the pgenesis directory, NOT pgenesis/src.
Also note that, unlike the case with serial GENESIS, 'make install'
or 'make nxinstall' is used to both compile and install PGENESIS.
This will create a shell script bin/pgenesis, which can be executed in order to start PVM and run the proper pgenesis binary file. The pgenesis binary is created in bin/<arch>, where <arch> is one of the architectures supported by PVM (e.g. ALPHA, LINUX, SGI5, SUN4, etc.) It will also create a .psimrc file that should be copied to your home directory.
If you find that you need to make changes in the Makefile that are not described in the Makefile.dist comments, or need to make other corrections in order to install and run pgenesis, please report them to us at genesis@genesis-sim.org.
After copying .simrc to your home directory, set your path to include pgenesis/bin, in order to be able to execute the pgenesis script from any directory. For example, if the Makefile setting for PARSRC_DIR is /usr/local/pgenesis and your shell is tcsh or csh, use:
set path = ($path /usr/local/pgenesis/bin)
If you are using bash as your shell, use
PATH=$PATH:/usr/local/pgenesis/bin
You can do a simple test of PGENESIS by running the demo.g simulation
script in the pgenesis Scripts/orient1 directory. After changing to this
directory, start it with "pgenesis demo.g". This is a 3-way parallel
decomposition of the genesis orient_tut simulation over 3 nodes. The
default is to create these nodes on a single processor. (Of course, to
achieve any advantage of using PGENESIS, you will want to use networked
workstations or a multiple processor machine. Details for doing this are
given in the documentation.)
The display will be somewhat like the serial GENESIS orient_tut simulation, and clicking on "sweep_horiz" should produce a similar pattern of activity and plots. A log file is created in o.out, so you will need to have write permission in the directory that contains this simulation.
For an even simpler test, or for debugging, you can try Scripts/hello.g.
Each user should be make sure that the PATH is set to include ~pgenesis/bin, where ~pgenesis is the root of the PGENESIS installation, i.e., the INST_DIR variable in the top level Makefile. If the script files you wish to use are not in your home directory, the examples directory, or the directory in which you run parallel GENESIS, you will need to add their directory path to the SIMPATH variable in the .simrc file you use. By default this is .psimrc for PGENESIS worker nodes.
You can run PGENESIS by typing "pgenesis scriptfile".
There are several available options, which may be listed by typing "pgenesis -help". The most commonly used of these are:
-config <file> where <file> is a file containing a list
of machines to be used (1 per line)
-nox run without Xodus
-debug tty create a window for each pgenesis node
-browse browse the documentation using netscape
For any use of PGENESIS beyond the simple test described below, see the documentation in the Doc and Hyperdoc directories. You can access the HTML documentation by pointing your browser at Hyperdoc/welcome.html. The README.mp file gives instructions for use on multiprocessor machines. The documentation also describes the use of PGENESIS on a network of workstations. Please report any inaccuracies to genesis@genesis-sim.org.
Normally the master node sends its output to your screen and the worker nodes send their output to the file /tmp/pvml.???? on the machine they are running on, where ???? is your uid on that machine. Many script bugs can be tracked down by examining those /tmp/pvml.???? files. Note that PVM appends to these files on every run, until PVM is halted and restarted. Also note that if a worker node exits without executing "paroff", some of its output to stdout/stderr which is buffered may not appear in the /tmp/pvml.???? file.
Output from the workers can be redirected to a file by giving the paron command the -output <filename> option. <filename> can be /dev/tty, in which case the worker output appears on your screen mixed with the master's output.
If you are running X-Windows, you can have each worker create its own window and send its output there by using the "-debug tty" flag when invoking PGENESIS. This requires that your DISPLAY environment variable is set to "hostname:0" when you invoke the pgenesis script.
By default PGENESIS starts the workers using -silent 3 which minimizes the messages printed by the workers (e.g., the banner). To see more worker status messages, give the paron command the -silent 0 option.
Common errors include:
0. User impatience leads to interruption of PGENESIS
before it has completed the initialization of the workers.
The timeout is set to 120 seconds, for maximum information
about the problem, let PGENESIS timeout if it fails
to initialize workers properly (i.e., wait!).
1. PVM can't find the worker executable. This is specified
with the -executable <filename> option of the paron
command, the default is "pnxgenesis". This executable,
or a link to it, should be in ~/pvm3/bin/$PVM_ARCH.
A symptom of this problem is that PGENESIS reports it
failed to spawn workers and the /tmp/pvml.??? files are
empty apart from the header.
2. PVM aborts worker startup due to I/O in your .cshrc
Symptoms of this are that PGENESIS reports workers
failed to spawn and the /tmp/pvml.??? files contain
error messages about TTY or other strange things, and
no GENESIS startup banner. Modify your .cshrc as
described in pvm-cshrc.stub.
3. The workers can't find the script containing the paron
command. Symptoms of this are that PGENESIS reports
workers failed to spawn and the /tmp/pvml.??? files
show the GENESIS prompt with no attempt to execute
paron, or the startup banners show that the script file
could not be found. Edit your ~/.psimrc and ~/.nxpsimrc
so that SIMPATH lists the directories containing the
scripts.
4. Incorrect use of barriers. Symptoms are that the master
or some worker(s) hang and eventually timeout. Examine
your use of barriers. Note that some commands contain
implicit barriers, it is possible there is a bug in
the source - if you suspect this mail ngoddard@psc.edu.
For more information on the finer points of using PVM, refer to the PVM user guide.
This section is intended for GENESIS users who have written or are using custom GENESIS libraries written in C that need to get compiled and linked into the GENESIS executable. For PGENESIS libraries are constructed just as for serial GENESIS, each in its own subdirectory in a canonical form. See the GENESIS documentation for descriptions of this form. In this section we describe how to link these libraries with PGENESIS components to create an extended PGENESIS.
Assuming your libraries are subdirectories of a root directory I'll call MYPGEN, these are the steps you follow to create and run extended PGENESIS.
'make nxextended' to make extended PGENESIS without Xodus
'make extended' to make extended PGENESIS with Xodus.
When using PVM-based PGENESIS on various SMPs:
On symmetric multiprocessors (e.g, Solaris multiprocessor, SGI Challenge, DEC AXP/8k series) you may need to increase the number of shared memory segments and perhaps other parameters of the shared memory system to run PVM effectively. For example, on Solaris multiprocessor, you need to add the following to /etc/system:
set shmsys:shminfo_shmseg = 100
set shmsys:shminfo_shmmni = 100
On ALPHA and HP SMPs, we have found that PVM does not always release the shared memory segments and semaphores it uses. Unreleased items can use up all available shared memory/semaphore resource. The script ~pgenesis/bin/ipclean attempts to release these resources. You can use the command "ipcs" to see the status of the shared memory system. In extreme cases a reboot is necessary to clean it out.
When using PVM-based PGENESIS on SGI/Cray Origin 2000:
There is a problem with PVM 3.4.beta3, PVM 3.3.11, and probably earlier versions, in which the pvm_trecv call consumes processor time while it is waiting. The effect that this has is to cause all PGENESIS processes on the Origin 2000 to continuously consume cycles, even if they are waiting for messages to arrive from elsewhere. So, if you have more processes than processors, this will cause a significant deterioration in performance.
When using PVM-based PGENESIS on DEC 8400:
PVM appears to have problems in occasionally generating unkillable processes. Machine reboot has been the only recourse at times.
On HPUX SMPs:
On HP-UX B.10.20 on an HP 9000/879 SMP, we found that serial GENESIS must be compiled with bison/flex:
YACC = bison -y
PARSER = bison
LEX = flex -l
LEXLIB = -lfl -lPW
LIBS = $(LEXLIB) -lm
XINCLUDE=/usr/include/X11R6
XLIB = /usr/lib/X11R6