Good rules

  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.

http://www.ibm.com/developerworks/linux/library/l-11sysadtips/index.html?ca=dgr-twtrLazyClusterAdmin&S_TACT=105AGY83&S_CMP=TWDW