CS 417 Exam 1

Spring 2007

    Part I – 44 Points

  1. 9 points
    As more and more people got access to the Internet, we weren't really running out of IP addresses in the 1990s even though most everyone said we were.
    (a) What was happening?
    (b) How did CIDR help?
    (c) How did NAT help?
  2. 5 points
    What is the purpose of the server-side stub in remote procedure calls?
  3. 6 points
    How did DCE RPC offer more flexibility in client-server applications than Sun (ONC) RPC?
  4. 6 points
    What two mechanisms does Microsoft ORPC/DCOM use to ensure that unreferenced remote objects are cleaned up?
    1.
    2.
  5. 6 points
    CIFS supports a SEEK command. NFS does not. Is this a deficiency in NFS? Explain.
  6. 6 points
    What's the major reason that AFS can scale to support a huge number of clients?
  7. 6 points
    Unlike NFS, AFS is a stateful file system. Explain the kind of state that the server needs to maintain.
  8. 6 points [OPTIONAL – EXTRA CREDIT] You create a directory on an NFS system and get an error that the file already exists even though it didn't exist earlier and nobody created one in the interim. When you check, the directory you created does indeed exist. Explain what might have happened.

    PART II – 56 points

    For each statement, select the most appropriate answer.

  9. Intel's Tera-scale processor prototype consists of 80 cores. Intel proposes a "3D die stack" memory interface, where each core has 256 Mbits of SRAM attached directly to it and other cores can access it via a 2D mesh interconnect network. This interconnect is best categorized as:
    (a) bus.
    (b) crossbar.
    (c) omega network.
    (d) NUMA (non-uniform memory access).
  10. Unlike a simple write-through cache, a snoopy cache:
    (a) ensures that cache contents remain coherent with main memory.
    (b) ensures that main memory is coherent with cache contents.
    (c) can work on a switched interconnect.
    (d) is faster for both read and write operations.
  11. A single system image is:
    (a) software that makes a collection of machines appear as one system.
    (b) software that uses distributed mutual exclusion to run on only one machine at a time.
    (c) a hardware interconnect to allow any processor to communicate with any other one.
    (d) tightly coupled software running on tightly coupled hardware.
  12. Ethernet sends a packet onto the network by:
    (a) waiting for the network to be clear and then checking that the it went out without a collision.
    (b) waiting for a predetermined timeslot to ensure that no other node transmits concurrently and causes a collision.
    (c) transmitting on a unique frequency band that no other node is using for transmission.
    (d) transmitting immediately and then checking that it went out without a collision.
  13. Which layer of the OSI protocol stack is responsible for managing the movement of data from one application to another?
    (a) Data Link (2).
    (b) Network (3).
    (c) Transport (4).
    (d) Presentation (5).
  14. Transferring an IP packet on an Ethernet network relies on:
    (a) IP-capable ethernet controllers.
    (b) nothing; IP and ethernet packets can coexist on the same network.
    (c) packet encapsulation.
    (d) CSMA/CD.
  15. Which of the following is not an attempt to help with managing efficient use of the network?
    (a) Flow detection.
    (b) Nagle's algorithm.
    (c) Differentiated services.
    (d) TCP packet retransmission.
  16. Which of these is the most likely scenario when contacting a DNS server to look up an IP domain name?
    (a) The server forwards the request to a root name sever.
    (b) The server returns the address of one or more root servers.
    (c) The server forwards the request to all other connected servers.
    (d) The server returns the address of another server that may know the answer.
  17. The part of RPC (remote procedure calls) that is responsible for converting between the network and native machine formats of the data being sent is part of this OSI layer:
    (a) Transport (4).
    (b) Session (5).
    (c) Presentation (6).
    (d) Application (7).
  18. Which of the following RPC systems does not require a separate interface definition language?
    (a) Sun (ONC) RPC.
    (b) DCE RPC.
    (c) Microsoft ORPC.
    (d) Java RMI.
  19. How does Microsoft ORPC/DCOM handle cleaning up unreferenced objects on the server if reference counting fails?
    (a) All remaining unreferenced objects are cleaned up when the client exists.
    (b) All remaining unreferenced objects are cleaned up when the server process exits.
    (c) The server periodically pings the client to see if it is still alive.
    (d) The client periodically pings the server.
  20. Microsoft ORPC enhanced DCE RPC with:
    (a) an interface-definition language.
    (b) use of XML for marshaling data.
    (c) a standard network data representation to support different architectures.
    (d) support for object/interface identifiers.
  21. The purpose of Microsoft's oplocks (opportunistic locks) is to:
    (a) allow distributed processes to negotiate a mandatory file and record locks on a remote file.
    (b) allow distributed processes to request an advisory file and record locks on a remote file.
    (c) allow the first process that accesses a remote file to get a lock for it.
    (d) tell the client what file data it can cache.
  22. Berkeley's xFS distributed file system project is unique in that it:
    (a) uses an upload/download model.
    (b) does not use TCP for connectivity.
    (c) gives the user a choice between session and sequential semantics.
    (d) does not use on a central file server.