CSE 121: Operating Systems, Architecture and Implementation
Fall 2003
Final Exam Review Session

Announcements:
 Final Exam on Tuesday, 12/09, 11:30-2:30 (SEQUO 147)

Note: these notes aren't complete, they're just going to give you an idea of
what to review. We went over a few of these items in a little more detail
in the review session.

(1) CSE 120 Review:
-------------------
     Address Spaces
     MMU 
     TLB, Virtual->Physical Page Tables
     Kernel/User modes
     system calls
     Processes
     example: shell ( fork(), exec() )

(2) File Systems:
-----------------

   - Disk abstraction / interface

   - example: DOS FAT File System

   + Fast File System (FFS):
     What did they improve, why did they do it, how did it improve it?

   + Improvements to FFS:
     extents, realloc, frag, co-locating, LFS (maybe), Journaling, Soft Updates
     (see handout)

   + Log-Structured File System (LFS):
     - Roughly, how did it work? How is it different from FFS?
     - Why did they believe LFS would improve FFS?
     - When is it better than and when is it worse than FFS?
     - How did they measure performance (specifically, this is difficult when
       you factor in cleaning overhead)?

   + Journaling versus Soft Updates:
     - How are Journaling file systems different from LFS and FFS?
     - What different types of journaling fs were measured in the paper?
     - How is Soft Updates different from Journaling?
     - Why would Soft Updates perform better than Journaling? When or why would
       it not (e.g. what are potential problems with it)?
     - Define what a rollback is.
     - Do the graphs/tables in the paper match up to your intuition? Explain the
       shape of some of the graphs in the paper.

   + Rio File Cache:
     - What did they do?
     - Why is memory unsafe?
     - How did they protect memory?
     - How did they test Rio (i.e. software fault injection, flip bits, etc.)?
     - Why is this so hard? Why might the results they have be difficult to
       understand?

(3) IPC:
--------

   - Identify different types of UNIX IPC.
   - What is a signal handler? How are they implemented (understand HW on it)?
   - What are process groups and how do they relate to shells?
   - How are signals used in shells/


(4) Virtual Memory:
-------------------

   - Page replacement algorithms:
     Be able to run through LRU, OPT, FIFO, Clock

   + Converting a Swap-Based SYstem to do Paging...
     - How did they approximate LRU?
     - What is a working set?
     
   + Implementing Global Memory Management in a Workstation Cluster
     - How did they perform global page replacement? 
       o which replacement policy was this an approximation of?
       o how did it work?

   + Implementation and Performance of .... Prefetching and Caching
     - What is prefetching?
       o when is it good?
       o when is it bad?
     - What 4 heuristics / rules were followed in the paper and why?
     - How does LRU-SP work?
       o what were the invariants they wanted to maintain with LRU-SP?
       o why are the S and P necessary in LRU-SP?
     - How do they manage allowing applications to specify replacement and
       prefetching policies?
     

(5) Microkernels and Virtual Machines:
--------------------------------------

   - What are microkernels good for and what aren't they good for?

   + Mach
     Identify the 5 primitives of Mach
     How did they improve IPC?
     How did they maintain binary compatability for existing applications?

   + L4: On u-kernel Construction
     What was the motivation for L4?
     What were the goals of L4?
     
   + Exokernel: Exterminate All Operating System Abstractions
     What were 4 arguments they gave against the Operating System abstraction?
       
   - Which ideas and motivations are similar between VMMs, exokernels and 
     microkernels? What is different?

   - What is a VMM?
     What advantages does it have? What challenges are there in implementing it?

   + Disco: Running Commodity Operating Systems on Scalable Multiprocessors
     Why did they build Disco?
     How did they deal with the NUMA architecture?
     Explain the graphs.

   + Memory Resource Management in VMware ESX Server
     How was ESX different from Disco?
     How did ESX share pages?
     What was ballooning?