Tuning after imaging is relatively simple compared to most other OpenEdge database components. Properly configured AI will have minimal overhead on the vast majority of OpenEdge databases.
All of the suggestions below are based on the assumption that performance is your priority and not ease of use or maintenance. You can certainly take different approaches but realize you are making a performance tradeoff for ease of use.
AI block size
There are rarely any universal absolutes, but this is one of them. Always set your AI block size to 16K. You can change the AI block size by using the rfutil command.
rfutil YourDB -C aimage truncate -aiblocksize 16
AI buffers are used to allow more work to happen in memory before being written to disk. The number of AI buffers is determined by the -aibufs startup parameter and each buffer uses one AI block size of memory (-aiblocksize 16 and -aibufs 200 = 3,200K of memory).
You can monitor your database during peak times to determine if you could benefit from more AI buffers by using ProTop or promon to look at the busy buffer waits and no buffers available metrics. Note that some small number of buffer waits are likely to exist in high volume applications so the goal is not to eliminate them completely but to keep the numbers as low as possible.
Make sure that you are starting an AI writer (AIW) for your database. The AIW will help offload some of the AI related activity from the rest of the OpenEdge database clients. To enable an AIW for your database you can check the appropriate box in OpenEdge Explorer/Management or add the following command to your database startup script:
Use multiple fixed length AI extents and size each extent large enough to hold at least 15 minutes of AI activity. If you have AI enabled you can monitor your current peak activity to determine a reasonable size. If you don't have AI enabled then you can monitor peak BI activity to determine an appropriate estimate.
If you are manually archiving AI files make sure to use the extract option instead of copying the actual AI file. This allows OpenEdge to extract only the used blocks from the AI extent and eliminates any downside with larger AI extents.