Run the model on a parallel computer

Discussion of how to use ROMS on different regional and basin scale applications.

Moderators: arango, robertson

Post Reply
Message
Author
angelolemos
Posts: 34
Joined: Wed Mar 11, 2009 4:24 pm
Location: UFES

Run the model on a parallel computer

#1 Post by angelolemos » Fri Oct 23, 2009 1:44 pm

People,
I have an important and single question.
I have a cluster with 6 machines Sun Solares, each one with 4 processors. All together I have 24 processors.
master
node01
node02
node03
node04
node05
I'm logged in the master machine.
I used: export USE_MPI=on
in my build.bash file
The problem is that the command ./build.bash -j 24 isn't working. The following error appear:
grep: illegal option -- P
Usage: grep -hblcnsviw pattern file . . .

./build.bash : Unknown option [ 24 ]

Available Options:

-j [N] Compile in parallel using N CPUs
omit argument for all avaliable CPUs
-noclean Do not clean already compiled objects


When I tried to run in the serial mode (I set #export USE_MPI=on), the following error appeared:
make: Fatal error in reader: makefile, line 32: Unexpected end of line seen

Somebody can help me?

Regards!

User avatar
kate
Posts: 3700
Joined: Wed Jul 02, 2003 5:29 pm
Location: IMS/UAF, USA

Re: Run the model on a parallel computer

#2 Post by kate » Fri Oct 23, 2009 4:50 pm

There is creating a ROMS executable to be run in parallel and there is executing make in parallel. Can you set "export USE_MPI=on" and then simply run the build script without the -j option?

As for the error, line 32 of the makefile contains code to make sure the make you are using is new enough. What do you get with "make --version"? I get "GNU Make 3.81" - this is the recommended (maybe even required) version to be using.

angelolemos
Posts: 34
Joined: Wed Mar 11, 2009 4:24 pm
Location: UFES

Re: Run the model on a parallel computer

#3 Post by angelolemos » Fri Oct 23, 2009 5:19 pm

Thank you kate.
I got the makefile in the package from ROMS, when I did the svn command.
https://www.myroms.org/wiki/index.php/ROMS_UNSW2008
The file is in the directory src with the others: Atmosphere Compilers Data Lib makefile Master ROMS User Waves

Have I to install the new version "GNU Make 3.81" and keep the file makefile?
When I executed in my own machine (Debian system and gfortran compiler) with the option serial, it did.

thanks!

User avatar
kate
Posts: 3700
Joined: Wed Jul 02, 2003 5:29 pm
Location: IMS/UAF, USA

Re: Run the model on a parallel computer

#4 Post by kate » Fri Oct 23, 2009 7:27 pm

What do you get with "make --version"?
"make" comes with your Unix/Linux system, at least with the development tools installed. It isn't a part of ROMS. The build script invokes "make" for you, expecting it to be there and of a sufficiently modern version.

The makefile on the other hand comes with ROMS and contains the instructions for compiling ROMS. The build script is an optional layer to insulate the timid from the syntax of "make".

angelolemos
Posts: 34
Joined: Wed Mar 11, 2009 4:24 pm
Location: UFES

Re: Run the model on a parallel computer

#5 Post by angelolemos » Mon Oct 26, 2009 12:31 pm

Kate,
I found other error that you already saw.

/afs/posseidon.dern.ufes.br/usr/angelolemos/src# gmake
makefile:235: INCLUDING FILE /afs/posseidon.dern.ufes.br/usr/angelolemos/src/Compilers/make_macros.mk WHICH CONTAINS APPLICATION-DEPENDENT MAKE DEFINITIONS
gmake: *** virtual memory exhausted. Stop.

I'm using the version GNU make 3.81, as you said. I typed "top", but there isn't anything with a great memory.
Why appear this "virtual memory exhausted" ?
How I can figure out this problem?

User avatar
kate
Posts: 3700
Joined: Wed Jul 02, 2003 5:29 pm
Location: IMS/UAF, USA

Re: Run the model on a parallel computer

#6 Post by kate » Mon Oct 26, 2009 4:32 pm

This is plain ROMS straight from the repository with the UPWELLING problem? I'm afraid I no longer have access to a Sun Solaris system to try it on. A web search of that error gives other people solving it by upgrading to version 3.81, so that's a puzzle. What happens if you type "limit" or "ulimit -a"?

User avatar
robertson
Site Admin
Posts: 123
Joined: Wed Feb 26, 2003 3:12 pm
Location: IMCS, Rutgers University

Re: Run the model on a parallel computer

#7 Post by robertson » Mon Oct 26, 2009 4:38 pm

angelolemos wrote: grep: illegal option -- P
Usage: grep -hblcnsviw pattern file . . .

./build.bash : Unknown option [ 24 ]

Available Options:

-j [N] Compile in parallel using N CPUs
omit argument for all avaliable CPUs
-noclean Do not clean already compiled objects
This grep error is due to the fact that your default grep command is not gnu's version of grep. The -P option instructs grep to treat the pattern '^\d+$' as a perl style regular expression and is not available in the Solaris versions of grep.

Post Reply