Comparison of the papers:
[FFF99] discusses power-law relationships of the Internet (i.e. rank,
degree, number of pairs within h hops, and eigenvalues), while [BKMR00]
focuses on the world-wide-web (in-degree, out-degree, size of components
follow power-law distribution, or combination of power-law and Poisson for
points close to the origin - e.g. nodes with small in-degree). The second
paper tries to go one step further and present the actual topology - one
big strongly connected component (25% of nodes) and relationships between
the other nodes and this component.

Discussion:
[both papers]
- Why power-law distribution? What is so specific about it? Maybe it has
to do something with hierarchical structure or self-similarity. The
authors of [FFF99] try to answer a similar question (why any distribution
at all? see section 5.1) on intuitive level - that adding a new
node(s) triggers a set of changes until a stabilized state is achieved. We
are a bit skeptic about this explanation and further investigation in both
cases would shed some light onto the process.
- Could we take an advantage of power-law distributions, e.g. design
algorithms based on the distribution knowledge? There already are some
algorithms, a search algorithm that goes to the neighbor of highest degree
should be discussed in the next class.

[FFF99]
- Critique:
  - How exactly are eigenvalues related to the rest of the study, apart
from being another parameter characterizing graph families? In what sense
is the knowledge of eigenvalues helpful for applications? We don't know...
  - Definition of neighborhood (1) is off by factor 1/2. 

[BKMR00]
- Would other crawling techniques give possibly different outcome? We
decided that the outcome should be the same (or similar). However,
allegedly the intersection of outcomes of different crawling techniques is
surprisingly small.
- What are the reasons for using BFS instead of DFS when DFS gives us
stronger results faster? We don't know and the answer might be related to
the use of Altavista's crawling. Or might it be some unknown memory
constraints?
- Critique:
  - It is not clear how they deal with sub-pages, e.g. is cnn.com one
web-page or tons of web-pages?

Possible future directions:
- Study coefficients of power-law distributions (changing in time)
- See [both], discussion point #1 (read papers [1],[2] suggested by Matei
in evaluations section?)
- See [both], discussion point #2 - other algorithms?
- Other Internet parameters to study and test for power-law (or
other) distributions
- Relationships between parameters studied? Does power-law distribution of
degrees imply power-laws of other parameter(s)?