You should use the long reads option if the reads (or contigs) will usually contain or cover more than one gene. So, if the reads are less than 500bp or are amplicon sequences, then they would qualify as short reads.
The short read mode is faster, but doesn’t deal well with the situation different parts of the read aligns to different genes.
When using long reads, one should also make sure that the alignment program can deal with them. For example, when using DIAMOND, use DIAMOND’s long read mode.
The daa2rma program is very slow and will be reimplemented in the future. Meganization of a daa file is much faster than using daa2rma.