1. Review Session for Fourth Quiz Jehan-François Pâris Summer 2011

2. Blue File System

3.   According to the designers of the Blue System what are the two limitations of flash drives?

4. Blue File System   They can be lost.   It is hard to keep them synchronized

5. Blue File System   The Blue File System is said to have a dynamic storage hierarchy. What does it mean?

6. Blue File System   The ranking of the storage devices in the storage hierarchy depends on their states.   A disk that is powered down will have a lower priority in the hierarchy than the remote server   A disk that is powered up will have a higher priority than the same server

7. Blue File System   How does the Blue file system operate its device write queues?

8. Blue File System   It empties them when it flushes them to disk.   Much more could be said.

9. Blue File System   Explain how the Blue file system saves energy by aggregating writes to local disks.

10. Blue File System   Aggregating writes to local disks saves energy by amortizing disk power state transitions across multiple writes.

11. Blue File System   True or false: Most of the Blue FS functionality is handled by a user-level server.

12. Blue File System   True

13. Pergamum

14. Pergamum  What equipment failures can be corrected by intratome redundancy?

15. Pergamum  Irrecoverable read errors

16. Pergamum  What would be the main drawback of a Pergamum system having  Plenty of intratome redundancy but  No intertome redundancy?

17. Pergamum  It would not tolerate full disk failures

18. Pergamum  How do intradisk parity blocks contribute to reduce the power consumption of the system?

19. Pergamum  They allow the local recovery of bad blocks without having to power up other tomes

20. Pergamum  What are the two main functions of Pergamum digital signatures?  Where are they stored?  Why?

21. Pergamum  Their two main functions are  To verify the integrity of the tome’s contents  By exchanging them with other Pergamum tomes, to verify the integrity of distributed data.

22. Pergamum  Where are they stored?  Why?

23. Pergamum  They are stored in a small flash drive so they can be consulted without powering the tome’s hard drive.

24. Pergamum  What is disk scrubbing?

25. Pergamum  Disk scrubbing periodically verifies that a given range of disk blocks can be retrieved and reconstitutes the contents of the blocks that it could no access due to an irrecoverable read error.

26. Pergamum  Which feature of Pergamum reduces the need for frequent full-disk scrubs?

27. Pergamum  Pergamum intratome parity reduces the need for frequent disk scrubs as it provides an additional way to reconstitute the contents of the blocks that caused irrecoverable read errors.

28. Pergamum  How does Pergamum reconstitute data contained on a tome that failed?

29. Pergamum 1.Pergamum replaces the failed tome by a new tome 2.One after the other, each tome in the same parity stripe as the failed tome sends its contents to the new tome

30. Pergamum  Why?

31. Pergamum  To avoid powering up too many tomes at the same time

32. Pergamum  How does the system’s workload—and intended use(s)-- affect the tradeoffs to consider when deciding the right amount of intra-disk and inter-disk redundancy in a storage system?

33. Pergamum  Intra-disk redundancy saves energy in archival file systems because it allows local reconstruction of irrecoverable read errors  We might prefer using more inter-disk redundancy in conventional file systems as inter-disk redundancy protects data against both irrecoverable read errors and disk failures.

34. FARSITE

35. FARSITE   How does FARSITE store users’ secret keys?   Why?

36. FARSITE   FARSITE encrypts the secret keys of its users with a symmetric key derived from user password and stores them in a globally-readable directory.   It does it because these keys are typically too long to be memorized by the user.

37. FARSITE   What characterizes a Byzantine failure?

38. FARSITE 1. 1.The failed node keeps communicating with the other nodes 2. 2.We have no easy way to detect such a failed node

39. FARSITE   How does Farsite guarantee the availability and the integrity its directory data?

40. FARSITE   Farsite replicates directory and manage them through a Byzantine fault- tolerant protocol that ensures their integrity (as long as less than one third of the machines misbehave in any manner).

41. FARSITE  In addition to using a Byzantine agreement protocol in its directory host, which steps does Farsite take to protect user files against malicious behaviors by its file hosts?

42. FARSITE 1.File blocks are encrypted so that file hosts cannot access their contents. 2.File blocks are also replicated on different hosts so that a single file host cannot maliciously destroy a file. 3.Farsite ensures that all copies of a given file block will be spread over machines controlled by different owners.

43. FARSITE  You are to design a FARSITE file system that can tolerate two Byzantine failures.  What is the minimum number of members in each directory host?

44. FARSITE  Each directory host should have at least seven members

45. FARSITE  What is the minimum number of copies each data block should have?

46. FARSITE  Each data block should have at least  Each data block should have at least three copies

47. FARSITE  What is a Sybil attack?  How does Farsite protects itself against them?

48. FARSITE  A Sybill attack is an attack where one or more rogue nodes assume multiple identities.  To prevent that, Farsite requires each node entered the system to have a verifiable unique ID issued by a trusted authority

49. FARSITE  Which actions does FARSITE take when the owner of a file grants or revokes access to a given file?

50. FARSITE  When the owner of a file grants access to the file to another user, FARSITE encrypts a copy of the file key with the public key of the new user. When that access is revoked, FARSITE deletes that copy.

51. FARSITE  How is the effect of a revoke different of that of the same revoke on a conventional UNIX system?

52. FARSITE  The user whose has lost the right to access the file could still be able to read it if he/she has kept a copy of the file key on his/her own workstation.

53. FARSITE  What could FARSITE do to implement the semantics of a UNIX access right revocation?

54. FARSITE  It would require encrypting the file with a new key.

55. FARSITE  What does FARSITE to improve its less than stellar response time?  Hint: Answer has two parts

56. FARSITE  Files are cached for up to one week on the client machines  Farsite uses background—”lazy”— propagation of directory updates

57. Farsite  What is a lease?

58. Farsite  A lease is a time-limited contract between the file server and a client guaranteeing that the server will not accept any update for a given file or et of files during the duration of the lease without notifying first the client.  Typical lease durations are fairly short.

59. Zyzzyva

60. Zyzzyva  Why may a Zyzzyva replica sometimes store two checkpoints?

61. Zyzzyva  Zyzzyva replicas have two checkpoints whenever their latest checkpoint contains non-committed history. (That checkpoint is then called a tentative checkpoint.) As a result, the replica must keep its previous checkpoint until the new checkpoint becomes a committed checkpoint.

62. Zyzzyva  When does a Zyzzyva tentative checkpoint becomes a committed checkpoint ?

63. Zyzzyva  A checkpoint becomes a committed checkpoint as soon as all the history it contains has become committed history

64. Zyzzyva  What are the four exchanges of messages that occur during the gracious execution of the Zyzzyva Byzantine fault-tolerant protocol?

65. Zyzzyva  Client sends a message to primary replica  Primary replica sends a message to all secondary replicas.  Secondary replicas send a message to the client.  Client send a message to all replicas (not included in the paper's figures)

66. FAWN

67. FAWN  How is the FAWN datastore organized?

68. FAWN   As a log operating in append mode

69. FAWN  Why?

70. FAWN   Because flash memory performs sequential writes much faster than random writes

71. FAWN  What is the purpose of allocating several randomly selected virtual nodes to each FAWN node?

72. FAWN  To spread the workload of a failed physical node among several physical nodes

73. FAWN  Why do Pergamum and FAWN select very different CPUs for their nodes?

74. FAWN  The CPU of a Pergamum tome controls a hard drive that is likely to be powered down 90 to 95 percent of the time  Power savings are paramount  The CPU of a FAWN node controls a faster flash drive that is very frequently accessed  Emphasis is on the best power-to-wattage ratio

75. FAWN  Consider a variant of Fawn tailored to a workload with infrequent requests to a very large data set  How would that affect your choice of a storage device?

76. FAWN  We should store FAWN datastore on a disk drive as the capacity of the storage device becomes more important than its access times

77. FAWN  How would your choice affect the organization of the in-memory hash table—and the size of the main memory?

78. FAWN  We would need a bigger main memory:  Many more hash table entries  Each hash table entries should contain a much larger key fragment to minimize false positives  Disk reads are much more expensive than flash memory reads