CS 417 Exam 2

Spring 2005

    Part I - 40 Points

  1. 8 points
    Explain the two techniques that ethernet cards employ to support multicast.
    1.
    2.
  2. 8 points
    (a) How does read-ahead help improve remote file system performance? (b) When does read-ahead not improve remote file system performance?
  3. 5 points
    Explain what is meant by a stateless file system.
  4. 5 points
    Why may opening a file under Coda be less efficient than under AFS?
  5. 5 points
    Explain reliable multicast.
  6. 9 points
    When its clock reads 3:42:07.100, a client sends a time request message to a server and gets a response back at 3:42:07.125. The response contains a server timestamp of 3:42:07.125. (a) Using Cristian's algorithm, to what value does the client set its clock? (b) If the best-case round-trip time is 20 msec., what is the error of the synchronized clock? [N.B.: 1000 msec = 1 second] (c) If the clock was last synchronized 1000 seconds ago, what is the drift based on your synchronization (in milliseconds per second)?

    Part II - 60 Points

    For each statement, select the most appropriate answer.
  7. A serializable class in Java:
    (a) guarantees that an object is reference by only one thread at a time.
    (b) ensures that valid remote references can be created for that object.
    (c) allows remote object references to be registered with the rmiregistry name server.
    (d) ensures that the object can be marshaled into a network message.
  8. Compared to a remote procedure model, an upload/download model is more likely to:
    (a) provide session semantics.
    (b) use a stateless architecture.
    (c) use remote procedure calls.
    (d) be implemented as a separate file system type on the client.
  9. Which of the following is not a remote procedure call under NFS?
    (a) write (write to a file)
    (b) close (close a file)
    (c) mkdir (create a directory)
    (d) statfs (get information for the file system)
  10. Which statement is not accurate about AFS?
    (a) AFS provides session semantics.
    (b) AFS volumes can issue referrals when a resource moves.
    (c) AFS does not consume disk space on the client.
    (d) AFS resources are allocated in units called volumes.
  11. Which statement is not accurate about Coda?
    (a) Replicated servers constitute a volume storage group.
    (b) Coda supports session semantics.
    (c) File updates are logged to a client modification log when a client is disconnected.
    (d) Replicated servers initiate a resolution process to synchronize their contents.
  12. Tokens in DFS (AFS version 3):
    (a) allow the server to control caching behavior on clients.
    (b) represent resources and are circulated from machine to machine.
    (c) allow a machine to ensure that all other machines are alive.
    (d) provide a way to meter traffic to the server.
  13. A persistent write-ahead log is important for:
    (a) providing an audit trail.
    (b) optimizing remote file system access.
    (c) ensuring that a transaction remembers its state even upon reboot.
    (d) allowing a client to play back file operations upon reconnection.
  14. The wait-die algorithm:
    (a) detects deadlock by having nodes send resource usage information to a coordinator.
    (b) detects deadlock by sending a probe message prior to requesting a resource.
    (c) does not allow a younger process to take resources held by an older one.
    (d) allows a process to wait for a resource to become free and then kill itself if the resource does not become available.
  15. Lamport timestamps can easily be modified to:
    (a) provide global ordering.
    (b) provide total ordering.
    (c) detect concurrency among events.
    (d) provide proper ordering among causally related events.
  16. Which mutual exclusion algorithm does not require group members to know all the members of the group?
    (a) Centralized
    (b) Token ring
    (c) Ricart & Agrawala
    (d) Lamport
  17. What is an efficient and scalable way for a node to find all cached copies of a given page in a distributed shared memory system?
    (a) Centralized directory
    (b) Distributed directory
    (c) Multicast page query
    (d) Broadcast page query
  18. Entry consistency:
    (a) is the opposite of release consistency: changes are propagated in the acquire phase.
    (b) propagates invalidations and updates anytime after the release phase.
    (c) cannot use the memory management unit effectively to detect changes to memory.
    (d) must be implemented for release consistency to work properly.
  19. A more practical approach to global time ordering is:
    (a) sync ordering.
    (b) total ordering.
    (c) causal ordering.
    (d) local ordering.