chameleon keeping data safe for the na ve and thri6y
play

Chameleon: Keeping data safe for the nave and thri6y Ansley Post - PowerPoint PPT Presentation

Chameleon: Keeping data safe for the nave and thri6y Ansley Post and Peter Druschel MPISWS NOTE: Slides contain notes which include addiGonal informaGon for readers of the slides ApplicaGon Scenario Home Users / Small Businesses


  1. Chameleon: Keeping data safe for the naïve and thri6y Ansley Post and Peter Druschel MPI‐SWS NOTE: Slides contain notes which include addiGonal informaGon for readers of the slides

  2. ApplicaGon Scenario • Home Users / Small Businesses

  3. ApplicaGon Scenario • Home Users / Small Businesses Is my data safe?

  4. ApplicaGon Scenario • Home Users / Small Businesses Is my data safe? What is RAID?

  5. ApplicaGon Scenario • Home Users / Small Businesses Is my data safe? What is RAID? How much is this going to cost?

  6. ApplicaGon Scenario • Home Users / Small Businesses Is my data safe? What is RAID? How much is Am I capable of this going to managing this? cost?

  7. Example

  8. Example Many possible ways to replicate, place and encode data

  9. Example Many possible ways to replicate, place and encode data

  10. Example Many possible ways to replicate, place and encode data

  11. Example Many possible ways to replicate, place and encode data

  12. Example Many possible ways to replicate, place and encode data

  13. Example Many possible ways to replicate, place and encode data

  14. Example Many possible ways to replicate, place and encode data

  15. Example Many possible ways to replicate, place and encode data

  16. Offline storage • Ideally, data is stored on at least one offline device – Enables recovery from catastrophic failure NOTE: All online copies may become corrupted due to virus, operator error, so6ware bug. Offline copies provide insurance against catastrophic failure.

  17. Offline storage • Ideally, data is stored on at least one offline device – Enables recovery from catastrophic failure NOTE: All online copies may become corrupted due to virus, operator error, so6ware bug. Offline copies provide insurance against catastrophic failure.

  18. AnG‐availability protocol • Writes are only allowed if a certain number of devices are offline – Enforced by a small trusted kernel NOTE: For clarity, a machine is shown as offline, but in pracGce only a storage device must be offline.

  19. AnG‐availability protocol • Writes are only allowed if a certain number of devices are offline – Enforced by a small trusted kernel NOTE: For clarity, a machine is shown as offline, but in pracGce only a storage device must be offline.

  20. AnG‐availability protocol • Writes are only allowed if a certain number of devices are offline – Enforced by a small trusted kernel NOTE: For clarity, a machine is shown as offline, but in pracGce only a storage device must be offline.

  21. AnG‐availability protocol • Writes are only allowed if a certain number of devices are offline – Enforced by a small trusted kernel NOTE: For clarity, a machine is shown as offline, but in pracGce only a storage device must be offline.

  22. AnG‐availability protocol • Writes are only allowed if a certain number of devices are offline – Enforced by a small trusted kernel NOTE: For clarity, a machine is shown as offline, but in pracGce only a storage device must be offline.

  23. AnG‐availability protocol • Writes are only allowed if a certain number of devices are offline – Enforced by a small trusted kernel NOTE: For clarity, a machine is shown as offline, but in pracGce only a storage device must be offline.

  24. AdapGve data storage • Use linear programming to select and adapt storage configuraGon Encoding ReplicaGon Placement NOTE: Huge number of possible tradeoffs. Well defined objecGve allows automaGc selecGon of best configuraGon.

  25. AdapGve data storage • Use linear programming to select and adapt storage configuraGon Encoding Offline ReplicaGon Placement NOTE: Huge number of possible tradeoffs. Well defined objecGve allows automaGc selecGon of best configuraGon.

  26. AdapGve data storage • Use linear programming to select and adapt storage configuraGon Encoding Offline ReplicaGon Placement NOTE: Huge number of possible tradeoffs. Well defined objecGve allows automaGc selecGon of best configuraGon.

  27. Conclusion • Currently finalizing design • Prototype, named Chameleon , is under development Contact: abpost@mpi‐sws.org for more informaGon

Recommend


More recommend