Duplicate data to avoid joins, use serializable transactions to update all the duplicated data.