use only fully overlapped reads?

I read the blog: http://blog.mothur.org/2014/09/11/Why-such-a-large-distance-matrix/
If I understand it correctly, we should use fully overlapped reads so that each base is sequenced twice. Thus, the sequencing experiment should be designed to generate mainly such reads, ie, the length of fragments should be chosen to be shorter than the read length. Is my understanding correct?

Correct.

Thanks. However, my goal here is only to discover abundance of KNOWN species listed in say Silva taxonomy. I have no need to do OTU analysis. Therefore, I believe I do not care of having too many unique reads, and I can tolerate more sequencing errors than OTU analysis. In my situation, I should have non-overlap paired reads to make longer reads so that they are able to uniquely map to a reference in Silva. Is my understanding correct?

Sure, but keep in mind that if you think you can get to “species” then every base of the variable regions is important. If you have a higher error rate then the odds are good that you will have at least one error in every sequence.

Pat