ABSTRACT
When moving to large processor counts, two scaling approaches may be used. Strong scaling is the case where the problem to be solved is set, and as the number of processors increases, the amount of work performed by each processor decreases proportionally. On the other hand, weak scaling sets the amount of work per processor constant, and as the number of processors increases, the amount of total work increases as well. Weak scaling is easier to scale, since the amount of computation performed by
each processor is constant and that help hide the communication. In weak scaling cases, the increased work is typically in the form of a ner or larger grid in a nite-dierence application or more particles in a particle in cell (PIC) application. It is important to only increase the work in the problem when it benets the science being delivered. If the increased work is not benecial to the scsience, then one must stay with the problem size that delivers the maximum understanding of the problem being solved and proceed in a strong scaling mode.
