[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] any suggestion on random no. seed for CONDOR run

On Thu, May 18, 2006 at 10:51:03AM +0800, Kwan Wing Keung wrote:
> Dear all,
> I would like to convert a Fortran program on massive statistical 
> simulation (originally running in a LINUX cluster using embarrassingly 
> parallel execution technique) to be deployed on a farm of student PCs 
> running Windows XP in a laboratory.
> The program can be easily converted I believe, however I would like
> to seek for any expert advise on assigning a seed for random no. 
> generation, for each program running in individual PC.
> I searched on the environment variables in the PC (through typing "set")
> but could not find any appropriate.  Using the clock time, MAC address or 
> IP address can also be possible.  But it is possible to dig out the 
> $(cluster).$(Process) CONDOR value within the program running in 
> individual executing PC?

I think you're best off generating random numbers in advance, and feeding them
in as commandline options. That way, if you ever need to rerun a job, you know the
seed that was used, and if there are ever any questions about the quality of your
random numbers your simulation generates you know where to start looking. 

If that's not important, you can always just pass $(Cluster) and $(process) as command
line arguments and use that.