For backups, as part of the restore procedure, there's a step that involves starting up a mongod on the data files in standalone mode to remove the previous replica set configuration and to optionally set the oplogTruncateAfterPoint document. Afterwards the standalone is restarted as a replica set node.
This switching of modes has a hidden cost during startup recovery. That is because all tables on standalones are logged and all tables on replicated collections are not logged (excluding the tables belonging to the local database).
Modifying the table log settings is an expensive operation. WiredTiger performs an fsync after each table is modified.
But because we're in standalone mode for a short period of time as part of the restore procedure, we should investigate if it's possible to skip changing the table log settings. This could speed up restores greatly when there's a large amount of WiredTiger tables present.