You need to set the -aiarcinterval parameter (in seconds). For example -aiarcinterval 900 will trigger an AI extent switch every fifteen minutes.
What is "good" isn't a technical question (which is what I tell my customers who ask the same thing
). It comes down to your RPO: Recovery Point Objective. In other words, in the event of a disaster, how much data can you afford to lose? It's about the value of your data to you; or in this case, the value of your client's data to them. That's a business question. Make them answer that question, in writing, and then set -aiarcinterval accordingly.
Be sure to set -aiarcdir to a local directory and then have an automated job to move the archived extents over to DR. It may be tempting to ask the AIMD to write the archived files directly to a network share that maps to the DR box but don't do this. If the share goes stale, network is flaky, etc., then the daemon will block and you might have to restart your DB to recover.