Hi @feixiang1209,
A couple of things come to mind here. NovaSeq runs are typically massive in size and it looks like as though you are not utilizing dada2's capabilities in running your task in parallel, so I'm not surprised this is taking a very long time. You can assign multiple cores to this task this with the --p-n-threads flag. That being said, you probably don't want to be running data from newer Illumina machines such as the NovaSeq with the q2-dada plugin. There's a bit of this topic discussed here and here. On the DADA2 github page it is mentioned that the next release will be able to handle the new binned quality scores but not sure when that is scheduled for. See the other threads about ways of using Deblur instead for this task.