Let me assume you want to do an excessive high number of inserts, so that you simply want to just ignore all other kinds of operations just to get those inserts to complete, without problems.
First, have you completely ruled out other types of databases? There are systems like industry databases that cope very well with massive amounts of inserts, typically used to receive and store data from equipment that is measuring something in a factory environment. Oracle is a relational database, it might not be the right type of software for your needs.
Having said that, let's assume you can, or will, or should, use Oracle. The very first thing you need to do is to consider all the various types of data you need to make this assumption about. If they're all about the same kind of data, you need 1 table, and it need to be lean and mean regarding inserts.
The optimal way do that is to do the following:
- do not add any indexes on this table at all, if you need a primary key, that's the only index you want
- if you need to do reads against this table, consider having a shadow table with indexes that you do reads, lookups, and aggregates against. If this doesn't have to be up-to-the-millisecond updated, consider a periodic batch job to update it with data from the master table. This will disturb the master table with read-locks as little as possible
- Make sure your server has fast disks. Transactional write operations will typically involve the disk at some point, so make sure that's a small bottleneck as you can get.
- If your application is gathering data from many incoming sources, consider adding a layer in front of the database that will keep the number of concurrent connections and thus transactions to that table to a minimum. If you get a high number of write-locks on the same page for an oracle database, ultimately your performance will suffer.
If you can split up the data, consider splitting it in such a way that it is stored on different physical disks. That way, disk I/O problems won't be cross-data-type, and only affect one type of data.
In the other end of the spectrum you have a denormalized table with lots of indexes optimized for a balance between lookups and updates, and you need to find some middle-way that will get you the performance you want.