CAM Load Balancing

[msl3v@virginia_edu]

Hello,

There seems to be a complex series of load balancing options available to CAM, though it's difficult to find descriptions or examples.

My specific issue, though, is how it may be possible to exceed the finite volume dycore's latitude-vertical decomposition process limit (e.g. 256 for 1.9 x 2.5 x 26).

There are some documents that pop up occasionally describing the use of namelist variables like phys_loadbalance or phys_alltoall that SEEM to allow physics or auxiliary processes to operate from an independent set of MPI processes... outside of dynamical limitations.

For example, http://www.iop.org/EJ/article/1742-6596/78/1/012082/jpconf7_78_012082.pdf
describes a set of 1.9x2.5 finite volume benchmarks at up to 1024 processes.
How is this configured?

Thanks.

ML

 

[rneale]

The paper documents the model allowing physics and dynamics to use a different number of processors which is not allowed in cam3.1 release code. They are not using independent sets of processors just that the physics will try and use all the processors you allocate at runtime and the dynamics will use its maximum limit of that is less than this value. unfortunately this code is only available in cam3.5/4 development code.