Index of /archive/p2psim/kingdata

Icon  Name                    Last modified      Size  Description
[PARENTDIR] Parent Directory                             -   
[   ] uking-t                 2004-06-29 15:19   23M  
[DIR] old-data/               2004-06-29 15:05    -   
[   ] mking-t                 2005-04-09 23:08   17M  
[   ] measurements.gz         2004-06-29 15:19  827M  
[   ] matrix                  2004-06-29 15:05   19M  
[   ] make_matrix_sorted.pl   2004-06-29 15:05  2.5K  
[   ] NS-list                 2004-06-29 15:05   72K  
[   ] IPs.gz                  2005-11-03 20:08  659K  

Note: use uking-t to run Vivaldi
use mking-t to run all other p2p protocols. 

The "King" data set.

This directory contains measurements of the latencies between a set of
DNS servers. It was used as the basis for evaluating the Vivaldi
network coordinate system among other distributed systems. 

A description of the king methodology can be found at:

http://www.cs.washington.edu/homes/gummadi/king/

Included are the raw measurements and a matrix and p2psim topology
file derived from that output. The derived matrix and topology file
are provided as a convenience. Those interested in, for instance, the
identities of nodes in the trace should produce a new matrix/topo-file
from the raw data.

This directory contains the following files:

* measurements.gz: 

The raw output of our asynchronous measurement tool, gzipped. This
file consists of one measurement per line. For example,

12.10.83.40 --> 12.104.199.65 1 101667 59192

indicates that a measurement was made from the host 12.10.83.40 to the
host 12.104.199.65. It was the first measurement in a series of five
(fourth column, see the above paper for an explanation of why multiple
measurements are necessary). The latency between the servers can be
found as the difference of the fifth and sixth columns: l = 101667 -
59192. The king method involves two DNS lookups: the fifth column is
the latency of a lookup to the first server (for a name that server is
responsible for), the sixth column is the latency of a DNS lookup
which is routed through the first server to the second. 

It may be convenient to read this file with the 'zcat' command. It is
approximately 4 GB uncompressed.

* make_matrix_sorted.pl:

A perl script which will generate a matrix from the measurements
file. It assumes that the input file is sorted. Run it as follows:

./make_matrix_sorted.pl <input file>

it will also read a file named "ignore" in the working directory. This
file contains a list of IP addresses of nodes which should not be
included in the final matrix. This list is helpful when filtering the
output matrix.

* matrix:

This is the matrix that results from processing the above file. The
entry in the ith column, jth row is the result of a measurement from
host i to host j. Some nodes which produced poor measurements have
been removed from this matrix. Units are in usec. These measurements
correspond to RTT.

* t-king:

This is a topology file suitable for use with
p2psim. It is derived from matrix. http://pdos.lcs.mit.edu.ezproxy.canberra.edu.au/p2psim/
Note that t-king contains one-way latencies; also, p2psim uses '1'
based addressing. Node 1 in the t-king file corresponds to the first
row in the matrix, not the second.

* NS-list:

This is the list of nameservers used in the trial along with a name
that the server should be responsible for. These nameservers are
filtered to include only names for which all responsible nameservers
are likely to be on the same subnet. This list could be useful to
those who wish to make their own measurements.

README:

this file.