This manual page is intended to provide instructions on how to use the GNP code's basic features. Many complex features will not be discussed here. Refer to the source code if necessary.

After the archive gnp-00.tar.gz has been downloaded and unpacked, there will be four directories:

• code - Contains the C code that implements the GNP algorithms (files: gnp.[h,c], simplex_downhill.[h,c])
• config_files - Contains sample configuration files for GNP experiments (files: 15L-8D, 6L-5D-Full)
• data - Contains the global Internet distance data used in our IMW 2001 and INFOCOM 2002 papers (files: matrix, combined)
• data2 - Contains the Abilene Internet distance data used in our INFOCOM 2002 papers (files: matrix, combined)

To compile the code, simply run the make.pl Perl script in the code directory, or just run "gcc -O3 -lm -o gnp gnp.c simplex_downhill.c". The resulting executable is gnp.

The gnp program expects to find two distance data files matrix and combined in the execution directory. It takes 3 parameters: (1) the number of lines in matrix, (2) the number of lines in combined, and (3) an experiment configuration file. For example, using the global Internet distance data and one of the supplied experiment configuration files, you can run:

./gnp 19 869 15L-8D

There are two major modes of running GNP, controlled by the experiment configuration file. The first mode is the asymmetric mode, when you do not have symmetric distance data between hosts. This is the typical case in Internet experiments because the number of measurement hosts is usually small, but the number of targets (i.e. hosts that only need to respond to ping) can be made large fairly easily. This is why two separate distance data files are used. The supplied global Internet matrix file contains the symmetric distances among 19 measurement hosts, while the combined file contains the distances between the 19 measurement hosts and 869 targets (smaller data set for the Abilene data). The 15L-8D configuration file is an example of operating in this mode.

The second mode is the symmetric mode, when you have symmetric distance data between hosts. For example, if you have a simulated topology, then symmetric distance data can easily be obtained between all hosts in the simulated topology. In this mode, the combined file is essentially ignored (thus it may be empty), only the data in matrix is used for computations. The 6L-5D-Full configuration file is an example of operating in this mode, you can run:

./gnp 19 0 6L-5D-Full

Click here to see the explanation of the format of the matrices matrix and combined.

What each GNP experiment does is controlled by the configuration file. In the following, we will use the 15L-8D file as an example for discussion. 15L-8D specifies 3 GNP experiments that use different combinations of 15 Landmark hosts and the 8-dimensional Euclidean space. The content of the file is shown below. Click on each hyperlinked section to see its meaning:

11
0 8 0 0 2  1 0  500 2000 0.000001 3 3 1 1  0  0  1 1 0 0  0
3
15  1 2 3 5 6 8 9 10 11 12 13 14 15 16 17
15  0 2 3 5 6 8 9 10 11 12 13 14 15 16 17
15  0 1 3 5 6 8 9 10 11 12 13 14 15 16 17

After running gnp, many output data files will be generated. Here we provide an overview of these output data files, for more details about the data files, please refer to the source code. As an example, after running "./gnp 19 869 15L-8D", the following set of files are generated:

• model.15L-8D.* - Each file corresponds to one Landmark combination experiment specified in 15L-8D. Each file contains information about the 15 Landmark hosts in an experiment. At the end of each file, after "======", the IDs and coordinates of the Landmark hosts are printed.
• test.15L-8D.* - Each file corresponds to one Landmark combination experiment specified in 15L-8D. Each file contains information about the 4 hosts in matrix that are not chosen as Landmark hosts, they are treated as ordinary hosts. After each "======", the ID and coordinates of a host is printed.
• target.15L-8D.* - Each file corresponds to one Landmark combination experiment specified in 15L-8D. Each file contains information about the target hosts in combined. At the end of each file, after "======", the IP addresses and coordinates of the 869 target hosts are printed.
• gnp.15L-8D.*.* - Each file corresponds to one set of predicted distances between a non-Landmark measurement host and the target hosts. For example, gnp.15L-8D.1.18 corresponds to experiment 1 (the second experiment), contains the predicted distances between the non-Landmark measurement host 18 and the 869 targets. The format of the file is the IP address of a target followed by the predicted distance.

When the symmetric mode is used, e.g. after running "./gnp 19 0 6L-5D-Full", the set of files generated is a little bit different:

• model.6L-5D-Full.* - Same as above.
• remain.6L-5D-Full.* - The format of these files is similar to those of test.15L-8D.* above, these files contain information about the hosts in matrix that are not picked as Landmark hosts.
• measuredfi.6L-5D-Full.*.* - These files contain the measured distances (as supplied in matrix) between each non-Landmark host and all the other non-Landmark hosts. For example, measuredfi.6L-5D-Full.0.10 is from the first experiment (0), contains the measured distances between non-Landmark host 10 and all the other non-Landmark hosts in matrix. The format is non-Landmark host ID followed by the measured distance. These files are generated to make comparisons between predicted and measured distances easier.
• gnpfi.6L-5D-Full.*.* - These files are similar to the measuredfi.* files, except they contain the predicted distances.

With these output data files, the coordinates of all hosts in the experiments can be extracted, the predicted distances can also be directly compared against the measured distances to evaluation the prediction accuracy.