LRVM paper
----------

1) What are the disadvantages in using Camelot for Coda?

2) Justify inclusion and exclusion of the mechanisms in LRVM as compared
   to a full blown transactional system such as Camelot.

3) LRVM is implemented as a library to be linked with an application
   desiring its services.  What are the implications of this design
   approach?

4) LRVM lives above the operating system.  The VM subsystem and the
   LRVM subsystem are thus independent of each other.  What implications
   does this have on functionality and performance?  In particular, how
   does this decoupling show up for applications running on top of LRVM?

5) LRVM provides "bounded persistence".  Explain.

6) The authors claim that structuring the LRVM as a user-level library
   as opposed to a collection of tasks communicating via IPC led to better
   scalability as evidenced by lower CPU load for LRVM compared to Camelot. 
   While this may be true since Camelot is built on top of Mach, would you
   agree with this reasoning as ground truth? 
   [Hint: Recall some of the papers we read on OS structures.]

7) How would you justify that LRVM is lightweight?

8) Develop the data structures and pseudo code for performing incremental
   truncation.


Rio Vista paper
---------------

1) What are the sources of problems in computer systems that leads to
   failure?  How are these relaxed in Rio Vista?

2) The Vista RVM library is 700 lines of code as opposed to 10K lines for
   a comparable functionality implemented in LRVM.  Explain why.

3) Develop the data structures and pseudo code for the Vista library
   (begin transaction, set-range, end transaction).

Quicksilver
-----------

1) What is the origin of the Quicksilver name for the OS presented in this
   paper? :-)

2) Give examples with justifications of servers that fall into the
   "stateless", "volatile", and "recoverable" classes, respectively.

3) Distinguish between "owner" and "participant" of a transaction.  

4) Identify points of similarily between LRVM and Quicksilver.

5) Quicksilver does not abort a transaction immediately upon detecting
   failure.  Why?

6) Discuss with examples of situations that are appropriate for each type
   the four different commit protocols (one-phase immediate, one-phase
   standard, one-phase delayed, and two-phase).

7) Discuss how Quickilver deals with participants that may join a
   transaction that is in the process of committing.

8) How does Quicksilver manage its log?  What are the optimizations used
   to reduce latency for log writing and dealing with memory pressure
   of excessive log generation?