Administrative:
 -No class on 12/9 and 12/11.
 -Only 18 paper reviews are required (down from 20).
 -Papers from the cancelled classes may be reviewed.

About the presentations:
 -Figure 2-3 minutes per slide and 10 minutes for Q&A =>
  don't use more than ~10 slides
 -Tom prefers 1 speaker, as it may lead to more 
  continuity, but hand-offs are permissible.
 -A ~5 minute demo might be a good addition to your 
  presentation, but let Tom know that you are going to
  be doing one beforehand.

On the papers so far:
 -You are likely to have a good grant proposal at the 
  moment:  an approach, preliminary results, etc.  
 -In the next few weeks, you will be getting actual 
  results, and those results should change the direction 
  (and intent) of your paper.
 -You may have to "throw away" much of what you have 
  already done (and rewrite it), so that it fits the 
  intent of your paper (as supported by the results).
 -Thus:
   1.  Decide what the point of the paper is.
   2.  Write the paper to make that point.
   3.  Review the paper...remove anything that does NOT
       support the point you are trying to make.
 -In general, the related work you have done may be
  too thorough.  Make it emphasize the main focus of 
  your paper.
 -The best papers are those that convince others that 
  one of their assumptions is untrue.  To exploit this,
  find a hidden assumption and prove a contradiction.

Peer-to-peer Resource Discovery:
 -There are many possible solutions in this realm.  The
  goals of the project will affect the solution that is
  produced.
 -Example application areas:
   IP-address lookup
   file-sharing
   publish/subscribe system
   object location
   multiplayer games
 -Note that DNS (which maps a name to an IP) faces many 
  similar problems with compared to Freenet (which maps
  a name to data) and Chord (which maps a key to a 
  value).
 -However, DNS has different goals than Freenet and Chord.
  For example, it does not support searching.  Can you 
  find u.washington.edu?  Sure.  How about "all CS
  departments?"  (cs.*.edu)  No way.  We can query DNS, 
  but it can't do queries on its own.
 -What about other support?  Some applications might 
  need triggers -- notification that data has changed.  
  Others might be context-sensitive and require different
  functionality depending on context.  The goal of the 
  application will affect the solution chosen.

What is wrong with the systems we have?
 -Our question really is, "What are their goals and do 
  we need something different?"
  1.  Are they searchable?
  2.  Does performance depend on locality?
  3a. Do they support anonymity?
   b. Decentralization?
  4.  How reliable are they?  Do you get what you ask for?  
  5.  How secure are these systems?  Do they prevent DoS?  
      Are they affected by malicious users?
  6.  Do they have a sense of fairness or resource balancing?
  7.  How tolerant of faults are they?
 -No system can do all of the above well.  

Why not use DNS?
 -It's centralized, inefficient, and has weak coherence.  
  (There is no guarantee that you'll get out what you put 
  in.)
 -We need a scalable, decentralized system.

Chord:
 -Lookups are performed in log time using a hash.
 -hash(node)->key, hash(file)->key
 -Finger tables store the data necessary to traverse the ring 
  in log n steps.
  ex:  On a 16-node ring, each node x contains the successors
       for x+1, x+2, x+4, and x+8 in its finger table.  
       Fingertable[10]
        successor for 11 
	successor for 12
	successor for 14
	successor for 2 (18mod16)
 -For redundancy, store every file at k successors.
 -What's good about about this?  People can come and go, and 
  the work necessary for addition and deletion is small.
 -What's not so hot?  There is no defense against malicious 
  nodes or people who insert nodes to take over all the files.  
  There is no sense of locality or bandwidth.

Freenet:
 -Freenet makes a completely different set of choices. 
 -They don't discuss their algorithm.
 -Unlike chord, they favor an absolute DIStrust model.
 -Clearly designed for anonymity and distributivity.
 -It's not clear that it works well or that it defends against
  bogus data.