Balancing Resources for Domains
The Domain Decomposition option for large models requires that you provide resources for the distributed solve by adding machines to the distributed machine workgroup.
Because the domains should be roughly the same size, you should provide balanced resources. As an example of balancing the load, suppose that machine called Biggie has 256 GB and another called Smalls has 64 GB. If you want up to five domains, it makes sense to enter Biggie in the workgroup four times and Smalls once.
Under Analysis Configuration Options, you should set the RAM Limit % to the memory desired for each individual domain.
In the example above, a good choice for Mem_Limit_Soft would be 64 GB.
Remember:
- A machine can appear in the workgroup more than once. The user should use this capability to maintain a balanced load.
- The first machine in the workgroup is used to control domain iterations.
- Under the Analysis Configuration Machine tab, the Number of Cores is also set per domain.
In the example above, a machine named Biggie has 16 processors and a machine named Smalls has 8 processors. Four domains will go on Biggie, so Number of Processors should be set to 4. Four processors on Smalls will be unused.
Recommendation: it is more important to use memory efficiently than to use all the processors.