thanks Vishal. that was helpful.
I also have the following scenario.
I have a cluster A of 5 machines and a cluster B of 5 machines. These two
clusters are under the globus environment.
For cluster A ->lamboot is done for its 5 machines and similarly for cluster B->
lamboot is done for its 5 machines. So all machines have a lam daemon running
on it. The lamboots on clusters A and B are done separately.
In this scenario, can a process that is running on a machine in cluster A spawn
processes on a machine in cluster B.
thanks
Siva
Quoting Vishal Sahay <vsahay_at_[hidden]>:
> Siva --
>
> To be able to spawn tasks, you should have lam daemons running on the
> nodes on which you want to spawn. So effectively this means you should
> have lambooted with those nodes in a single scope.
>
> As I see, you can do this under two scenarios --
>
> * If your clusters are rsh based, that is it possible to access individual
> nodes across clusters via rsh/ssh, then you can include all the hostnames
> in the host file for lamboot (these hostnames may be across clusters)
>
> * If you are using the "globus" environment, then these clusters
> effectively form grid machines for you and should be able to use the LAM
> globus boot module for this purpose (check the globus boot section in the
> user doc for LAM -- http://www.lam-mpi.org/using/docs/ -- be sure to
> provide the proper contact strings with the certificates and hostnames in
> the host file for lamboot)
>
> For other boot modules, like PBS and BProc, LAM currently does not support
> lambooting acrosss clusters; or if you have clusters with different
> scheduling mechanisms (like PBS on one and BProc on other), you cannot
> currently lamboot across such clusters.
>
> Hope this helps --
>
> -Vishal
|
