Raymond Chen

Pretty old post, but still worth jotting down:

1. Don’t reinvent the wheel.
2. Use open source.
3. Automate everything.
4. Design for scale – plan to be lazy from the start.
5. Design for hardware manageability.
6. Use good cluster management software – don’t dig wells with teaspoons.
7. Use open source monitoring solutions.
8. Control your users with a queuing system.
9. Verify you get what you pay for – benchmark it!
10. Manage cluster communication.
11. Look for ways to become lazier.

Source