The 12.0.1 docs on Snapshot isolation contain the following statements:

When snapshot isolation is enabled for a database, each time a row is updated, the database server adds a copy of the original row to the version stored in the temporary file. The original row version entries are stored until all the active snapshot transactions complete that might need access to the original row values.

I have two related questions to this:

1. "...each time a row is updated..."
   Does that mean: Each time a commit is made - meaning if a transaction would change a row several times, only the last state would be stored as a versioned row?

2. "...are stored until all the active snapshot transactions complete..."
   What happens when there is no current transaction using any of the snapshot isolation levels at the time the row version is stored? Will it be stored anyway (and immediately deleted - or short after) or will the storing be skipped?

Background: I'm trying to figure out the impact of snapshot isolation when only now and then a transaction would use that mode, i.e. when updates would mostly take time without a concurrent snapshot isolation transaction - would those updates lead to possibly noticeable worse performance?