Currently PDF convolution can be parallelized over properties. This is a problem for many flamelet libraries with Arrhenius properties (likely...), as a load imbalanace leads to one property taking 5-10x longer than the rest of the table. A natural solution to this is to add parallelism - parallelize over both properties and the newly added convolution dimension (typically the scalar variance).
Currently PDF convolution can be parallelized over properties. This is a problem for many flamelet libraries with Arrhenius properties (likely...), as a load imbalanace leads to one property taking 5-10x longer than the rest of the table. A natural solution to this is to add parallelism - parallelize over both properties and the newly added convolution dimension (typically the scalar variance).