Categories: Cloud Posted by mheydt on 1/25/2009 11:02 PM | Comments (0)
  1. Package files for upload instead of uploading many small files
  2. Utilize multi-threaded access to blobs
  3. Utilize exponential back-off and retry when calling services
  4. Store reference information within messages
  5. Use Process-oriented messaging
  6. Use Document oriented messaging
  7. Use queues to distribute load
  8. Utilize controller systems to start and monitor clusters
  9. Use table based storage to store metadata, not the actual data
  10. Launch multiple instances at once
  11. Allocate compute nodes out of pools to lower startup times
  12. Automate as much as is possible
  13. Keep systems loosely coupled
  14. Think parallel
  15. Think on-demand
  16. Design for reboot and relaunch