Do as little as possible - Focus on simplyfing the processes instead of achiving a theoretical design perfection.
Avoid multiple IO to the database - bundle queries to a procedure and call thay in ur app. again in procedures reduce IO to the db.
Avoid multiple connections to the DB - Connecting to the db is the slowest process.
Store data the way u need it.
Go Atomic - Divide and conquer. Use partitions, Local Indexes.
Spread data across disks - Use different tablespaces for tables and indexes and partitions.
Use MVs - reduce operations performed on tables.
Use parallelism - almost everything can be parallelized.
Avoid the need to query undo segements. - It need to look multipe locations for a piece of data.
Keep stats updated.
Use Disk caching.
Use Larger block size. - keeps more data in the memory.
Make sure blocks are densly packed. - adjust pctfree. mind row chaining and migration.
Try to perform sorts in memory.
Use a cpu with SMALL no of FAST processors.
Use Standard initailization params.
Use proper use of redo logs, checkpoints.
Have sufficient rollback segments.
Small tables can the kept in the keep pool using the buffer_pool clause of create table.
Be free from Deployment and Migration Errors.