Unit One Objectives          CSCI 524       Distributed Operating System Design
Tobin Maginnis                              Updated 29-Jan-08

Lecture
 1)Contrast loosely versus tightly coupled architectures and give two
   types of tightly coupled designs seen in multi-core architectures.
 2)Describe the three basic design tradeoffs in distributed processing:
   a)Resource Sharing - Heterogeneity in form, function, & speed
   b)Reliability - Integrity & Availability through redundancy
   c)Performance - Linear speedup
   and explain how each of these goals are in conflict with one another.
 3)Define SMP and compare it to traditional multiprocessor design goals.
 4)Describe the Single System Image (SSI) effort and compare it to traditional
   multiprocessor design goals.
 5)Describe a generic Linux cluster and compare it to traditional multiprocessor
   design goals.

Experience Using Multiprocessors Systems - A Status Report
A. K.  Jones and P. Schwarz,  Computing Surveys, Vol. 12,    Nov. 2, 1980.
 1)Define uniprocessor, multiprocessors, computer networks, distributed
   computers, and array or vector computers.
 2)List the four parameters one must consider when evaluating a
   multiprocessor system and give three ways they interact.
 3)Describe three "dimensions" of parallel solutions.
 4)Describe five levels of "granularity of the computation unit."
 5)Define "locality" and "regularity" of reference and explain how they may
   help or hinder a multiprocessor environment.
 6)Explain why the asynchronous parallel PDE executed faster than the
   synchronous PDE.
 7)Define "linear speedup." Explain why certain types of searches can be
   performed faster than a linear speedup. Describe three problems this
   solution.
 8)Describe four possible techniques for constructing a parallel solution.
 9)Discuss the myth which states: "With each additional processor a more than
   additive increase in overhead prevents linear speedup of the application."
10)Describe C.mmp's physical configuration in PMS notation.
Omit)Contrast "objects" and "capabilities," and describe the significance of
   objects as they are employed in complex programming tasks.
12)Describe Cm* physical configuration in PMS notation and contrast its
   fundamental difference with C.mmp.
13)Explain the authors observation that "there are few differences between
   multiprogrammed systems with and without multiple processors."
14)Explain the tradeoff for employing homogeneous versus non-homogeneous and
   ditto for symmetric versus asymmetric architectures or I/O devices.
Omit)Support or refute the concept that granularity of computation interacts
   with various software engineering issues such as programming style, etc.
16)Explain which parts of the following statement are true and false: 
   "Parallel systems of programs are difficult to design, to build, and to
   understand."
17)Explain why or why not "software that is physically distributed must be
   more reliable than if it were physically centralized."
18)Describe the three aspects of resource management and how this issue
   affects the user-variable aspect of any computer system.
19)Explain four aspects of "memory reference locality" and how it interacts
   with multiprocessor architectures.  How might this process be automated?
Omit)Explain the "twist" on scheduling in multiprocessors and give three
   reasons why its seems to offer an advantage.
21)Rank and describe the six highest sources of multiprocessor performance
   perturbation in C.mmp.
Omit)Describe how the multiprocessor scheduling "twist" could be automated and
   explain when this is not necessary.
23)Describe the two dimensions to reliability and how they relate to recourse.
24)Describe "distribution of control" and explain why this appears to be
   difficult at best.
25)Contrast passive versus active errors and explain the confounding which
   active errors introduce into distributed authority.
26)Given the lack of past success with multiprocessor design, explain the
   strong interest in distributed operating system research.