Homework II
Due Date: December 3, 1997 (at start of class)

1. Remote procedure calls simplify the task of distributed programming
   because the automatically generated stubs take care of many of the 
   problems that application programmers must deal with when programming
   in a message passing system. Explain what these problems are and 
   how/why they can be handled by the stub code.

   The RPC design discussed in the Birrell/Nelson paper tries to provide
   at-most-once semantics despite communication and node failures. Explain
   what mechanisms are employed by it to deal with failures of server nodes.

2. The server side of a distributed file system implementation must
   receive  client requests and process them as they are
   received. Provide detail of the server side for a file system
   such as Sprite. In particular, list the type of messages received
   and sent by the server side, the major data structures maintained
   by it and the management of control (e.g., multithreading) at the 
   server side. For full credit, describe how requests are dispatched
   and handled.

3. Can a distributed file system (DFS) be implemented using a subsystem
   that provides consistent distributed shared memory (DSM)? If this is a
   good idea, identify what simplifications are possible in the DFS design
   and implementation because of the availability of the DSM system.
   If you do not recommend building DFS over a DSM system, explain why 
   this is not a good idea.

4. Does there exist a set of real-time tasks such that the tasks in the 
   set can be scheduled by the least-laxity algorithm but not by the 
   earliest-deadline algorithm? If your answer is yes, give 
   a set of tasks and values for the parameters that describe them 
   (e.g., period, execution time). If the answer is no, explain why
   such an example cannot be created.