CS 417 Exam 1

Spring 2009

    Part I – 40 Points

  1. 5 points
    What is the purpose of the snooping feature of a snoopy cache?
  2. 5 points
    Your network requires the ability to address 10,000 hosts. A class B network, with a 16-bit host address, has the ability to address 65,534 hosts. 55,534 potential addresses are hence wasted. Will using CIDR instead of a class B address help? Explain how or, if not, explain why not.
  3. 5 points
    How does NAT/PAT help us from running out of global IP addresses?
  4. 5 points
    What is the architectural difference between a NUMA architecture versus using cache memory in a traditional bus-based SMP system?
  5. 5 points
    Both ARP and DNS are examples of distributed name servers. Contrast the query approach between ARP and DNS (not what they do but how they do it).
  6. 5 points
    What is the purpose of a client stub function in RPC systems?
  7. 5 points
    Why might session semantics make a distributed file system easier to implement?
  8. 5 points
    How does the Google cluster architecture scale within a data center to support larger databases (i.e., an ever-increasing web) without using faster processors or disks? [The question is about optimizing the time a query takes, not about overall throughput, which can be increased by adding more web servers and replicating other servers.]
  9. PART II – 60 points – 3 points each

    For each statement, select the most appropriate answer. You may omit one question

  10. Graphics processing units (GPUs) are examples of this architecture:
    (a) SISD
    (b) SIMD
    (c) MISD
    (d) MIMD
  11. An SMP (Symmetric Multiprocessing System) is characterized by:
    (a) shared memory among multiple identical processors.
    (b) a switched network connecting several identical computers.
    (c) a bus-based network connecting several identical computers.
    (d) replicated processors and memory for fault tolerance.
  12. For an operating system to use a NUMA architecture effectively, it would not be a good idea to:
    (a) reschedule threads to run on the same processor that was used for that thread in the past.
    (b) attempt to allocate memory that is local to the processor on which the thread making the request is running.
    (c) distribute a thread's memory allocation requests throughout the entire memory space to spread the load.
    (d) perform I/O operations into memory associated with the processor (node) that initiated the request.
  13. A single system image refers to:
    (a) a collection of software that makes a bunch of independent computers appear as a single system.
    (b) a single distributed operating system that is run over a collection of computers.
    (c) system software that mirrors processes and disk resources onto one or more additional computers for fault tolerance.
    (d) system software that allows programs to automatically exploit the parallelism of a multiprocessor system.
  14. A virtual circuit is most likely to offer:
    (a) guaranteed bandwidth.
    (b) guaranteed minimum latency.
    (c) in-order packet delivery.
    (d) all of the above.
  15. Ethernet is an example of:
    (a) a reliable circuit-switched network.
    (b) an unreliable circuit-switched network.
    (c) a reliable packet-switched network.
    (d) an unreliable packet-switched network.
  16. Which layer of the OSI network stack does UDP/IP represent?
    (a) Data link (2)
    (b) Network (3)
    (c) Transport (4)
    (d) Presentation (5)
  17. Nagle’s algorithm:
    (a) allows a programmer to set bits in an IP header to request a certain level of service.
    (b) groups related flows of IP packets together at the router.
    (c) avoids sending lots of small TCP packets.
    (d) transmits IP packets at increasing intervals to relieve network congestion.
  18. What is the difference between hard and soft Quality of Service (QoS)?
    (a) Hard QoS guarantees a services level while soft QoS is a best-effort approach.
    (b) Hard QoS requires hardware support while soft QoS is a software-only solution.
    (c) Hard QoS only works within a local area network while soft QoS can work across the Internet.
    (d) Hard QoS relies on traffic policing while soft QoS relies on traffic shaping.
  19. ATM (Asynchronous Transfer Mode) was designed to:
    (a) address the delivery of real-time media as well as non-time-critical data transfers.
    (b) provide the ability to send data asynchronously without having to wait for acknowledgements from the remote system.
    (c) provide the ability for both sides to send data to each other concurrently.
    (d) be a networking technology that is considerably less expensive than IP.
  20. The Domain Name System is responsible for translating:
    (a) human-readable computer names to IP addresses.
    (b) IP addresses to ethernet MAC addresses.
    (c) ethernet MAC addresses to their corresponding IP addresses.
    (d) internal IP addresses to external IP addresses.
  21. A key design goal in Google's cluster architecture is:
    (a) maximizing energy efficiency & minimizing price/performance.
    (b) maximizing peak processor performance.
    (c) using commodity PCs.
    (d) maximizing disk throughput.
  22. A portmapper on an ONC (Sun) RPC system is a name server that:
    (a) associates client ports with server ports.
    (b) allocates available port numbers to RPC server stubs.
    (c) allows a client to request an unused port for communicating with the server.
    (d) allows the client to find out on what port an RPC server is listening.
  23. How does the cell in DCE RPC provide programmers with a greater amount of transparency?
    (a) It handles all the details of marshalling and unmarshalling data.
    (b) It allows programmers to use a unique universal ID (UUID) to refer to an interface.
    (c) It allows programmers to specify TCP/IP or UDP/IP communication at run-time.
    (d) It allows clients to find what machine a specific service is running on.
  24. Microsoft Object RPC (ORPC) is compatible with but differs from DCE RPC in that Microsoft's RPC:
    (a) added an Interface Definition Language.
    (b) added support for identifying remote objects.
    (c) does not require a precompiler.
    (d) moved from a binary encoding to XML.
  25. Microsoft DCOM's distributed garbage collection does not do the following:
    (a) Remote reference counting.
    (b) The client periodically sends a list of remote objects IDs in use to the server.
    (c) The server pings the client at regular intervals.
    (d) All of these are used in DCOM's garbage collection.
  26. Java RMI does not require:
    (a) a precompiler.
    (b) data conversion when marshaling to support different machine architectures.
    (c) remote garbage collection.
    (d) a name server.
  27. Which of the following is not a valid NFS function in the original version of NFS?
    (a) lookup: look up a file name and return a handle to it.
    (b) mkdir: create a directory.
    (c) readbytes: read bytes from a file.
    (d) close: close a file.
  28. SOAP does not help deal with:
    (a) locating machines hosting remote services.
    (b) creating requests that can cross firewalls more easily.
    (c) different data encodings due to different machine architectures and programming languages.
    (d) invoking methods on remote objects.
  29. TCP provides the following advantage over UDP:
    (a) It consumes fewer resources by not maintaining connection state.
    (b) It regulates the rate at which packets enter the network to reduce congestion.
    (c) It sends the source address and port number so the to identify the origin of the packet.
    (d) It provides a mechanism to retransmit missing or dropped packets.
  30. What is the primary use of a DNS referral?
    (a) Managing addresses of machines that have moved outside the domain.
    (b) Managing addresses of machines that have moved within the domain.
    (c) Translating between aliases of machine names and the actual name.
    (d) Identifying a name server that is closer to the requested domain.