Hello Tony,
this is highly dependent on what the data is needed for and how small a time step the model will be taking. For instance, I often run models with 1 us time steps down to 5 ns time steps. If I am looking at signals like my line frequency, I don't necessarily need that level of data, and will decimate to 1kHz (so decimmation of 1000 sometimes). For smaller runs, it isn't worth it to decimate, but as they get larger, it can be faster to process the data if you take less of it. You can always start with a decimation of 10, and see if it is good enough. If it isn't, go back to 1. If it is good enough, you can try 100. Basically, think about what level of resolution you need for a plot, vs what time step the simulation needs to converge.
Regards,
Joel
