{Mid Semester - I} [Teaching @ PUCSD] [Home]

Mid Semester Examination - I
(04 January 2000)

  1. Give brief/short answers for the following -

    1. Why do most RPC systems support "call by value" semantics for parameter passing?
    2. The object locating operation can be greatly simplified if the location of an object is embeded in its unique identifier. Inspite of this advantage, why is it undesirable to include the location of an object in its unique identifier?
    3. In the design of a distributed file system, high performance and high reliability are conflicting properties. Discuss.
    4. Concurrency transparency is a desirable goal for distributed systems. Do centralised systems have this property automatically?
    5. Under which circumstances would a client prefer a location transparent distributed file system (DFS) over a location independent DFS? Discuss the reasons for these preferences.
    6. Some distributed file systems use client caching with "delayed write-back to the server" or "write on close". In addition to the problems with the semantics, these systems introduce another problem. Explain, what is it?

  2. The operations performed by a server are non-idempotent. Describe a mechanism for implementing exactly-once IPC semantics in this case.

  3. What is an immutable file? Can a file system be designed to function correctly by using only immutable files? If no, explain why? If yes, explain how the basic file operations (create, write, read, delete) can be performed in this file system for shared files.

  4. In the design of a distributed operating system, the data caching mechanism may be used for caching many different types of data. A separate cache can be maintained for each type of data. In your opinion, is it necessary to always keep a cached data upto date? If yes, explain why? If no, give an example in which a system can function correctly even when processes access cached data that are not always upto date.

  5. In a certain workstation based system, the workstations have local disks that hold the system binaries. When a new binary is released, it is sent to each workstation. However, some workstations may be down (or switched off) when this happens. Devise an algorithm that allows the updating to be done automatically even though workstations are occasionally down.

  6. Process migration with a heterogeneous network is usually impossible, given the differences in architecture and operating systems. Describe a method for process migration across different architectures running

    1. same operating system
    2. different operating system

  7. Consider a distributed system with two sites, A and B. Discuss whether site A can distinguish among the following -

    1. B goes down.
    2. The link between A and B goes down.
    3. B is extremely overloaded and response time is 100 times longer than normal. What implications does your answer have for recovery in distributed systems?

Created on May 04, 2006