Are you already have answers on the last question? They are also are very interesting for me. Thanks for replies.
In that particular case, topic starter worked with v3-v4 region. Those amplicons are relatively large and it is preferable to sequence them as 300X300 paired reads. But it looks like library was sequenced as 250X250 paired reads and truncation resulted in insufficiency of overlapping bases and a lot of reads failed to merge.
So, one need to account for overlapping region to decide which truncating parameters are the best for the dataset in question.