Abilene

Collection methodology

The idea is to directly work with destination AS numbers (which
themselves are computed by the routers using the BGP tables when each
netflow record is generated).

(1) Examine the netflow leaving the egress links and build a set of
    destination AS numbers that egress at each router

(2) From (1), for each destination AS number, we can find all the egress
    routers correspond to this AS.

(3) At each ingress router, for each flow record, I can map its
    destination AS number to a set of egress routers using results
    of (2).  I then simulate OSPF to pick the one egress router
    that is the closest.

The above works as long as the destination AS is not Abilene (AS11537).

When the destination is Abilene, in principle you could do the same
thing on destination prefixes. Unfortunately, the Abilene netflow
mask out the last 11 bits of all the IP addresses, so it is impossible
to get the true destination prefixes for Abilene. As a result, the
TMs generated do not include any traffic destined for AS11537
(fortunately they don't account for too much traffic).

Scripts:
convert_data.pl - convert the data from its original format
into the format presented here

tm_totals.pl           - create totals for each TM
tm_totals.m	       - draw pictures of totals, and create cleaned data

tm_decomp.sh	       - do decomposition of the totals
		         uses decomposition code (see below)
tm_decomp.m	       - draw pictures of decomposed parts
tm_fourier_analysis.m  - simple Fourier analysis of 2 weeks of the data
tm_fourier_analysis2.m - simple Fourier analysis of 6 weeks of the data

write_tm_latex.pl      - write out a TM in LaTex format

Summary Data:
tm_totals.dat - total traffic per TM from "Measured" data
tm_totals.clean
tm_totals.clean.datenums
tm_decomp.dat

Data Directories:
Measured - original TM, from collected data, as
described above
SimpleGravity - simple gravity model based on Measured
SimpleTomoGravity - tomogravity results using simple
gravity prior (see [1])
GeneralGravity - generalized gravity model (see [1])
GeneralTomoGravity - tomogravity results using generalized
gravity prior (see [1])

External code/files: (in separate download)
TrafficMatrix.pm - has code for reading and writing files
README - README describing file format and
philosophy

decomposition	  - C code for time-series decomposition

Other notes:

In several places paths are hard-coded. They obviously have to be
fixed to install. This could be better, but there aren't too many of
them, and they are pretty obvious.

I haven't provided the tomogravity or synthesis code yet. We really
need to create new versions of these.

References:

[1] Fast Accurate Computation of Large-Scale IP Traffic Matrices
from Link Loads, Yin Zhang, Matthew Roughan, Nick Duffield and
Albert Greenberg, ACM SIGMETRICS 2003.