That is not a easy question and the answer is influenced by many parameters, to mention some:
- On-premise or cloud?
- How may app-engines if cloud?
- Database?
- What is 'Handled'?
- What is 'performance issue', longer than 3 seconds, 2 hours?
In general some recommendations
- When replication data from other systems: don't try to import >million records every day, optimize that
- Bulk data processing microflows: Work with batches/endTransaction (java-action) to prevent java heap space errors
- For big data retrieves: use OQL to get the database speed
- Import in intermediate tables and process from that point
- Use checksums or other mechanisms to prevent syncing twice the same data.
- Limit the number of indexes: they will cost more to maintain than they will gain.