HEDGE

Getting Started Instructions

Install the necessary pacakges with pip install -r requirements.txt.

We provide code and data to produce results from the main paper, for all of HEDGE-AGG and for the publicly available topologies for HEDGE-TE. To use the HEDGE-AGG optimization algorithm, execute a Python script like this to print the results:

from wavelength_aggregation import capacity_provisioning_alg

NUM_WAVELENGTHS = 50
MAX_CAPACITY = 5000  # in Gbps
MIN_CAPACITY = 3000  # in Gbps
AVAILABILITY_TARGET = 0.991  # e.g., 99.1% availability

# Probability of a wavelength operating at each data rate (s_i in the paper) given the SNR
capacity_dist = {250: 0.99, 0: 0.005, 150: 0.004, 200: 0.001}

# Returns the number of wavelengths to operate at each data rate (i.e., modulation format)
counts = capacity_provisioning_alg(capacity_dist, NUM_WAVELENGTHS, MAX_CAPACITY, MIN_CAPACITY, target)

To use the HEDGE-TE optimization algorithm, first create a pickle file with the stochastic topology, a dict with format {<directed_edge>: {<capacity1>: <prob1>, <capacity2>, <prob2>,...}, ...} where the <directed_edge> key is a (str, str) tuple for a WAN link and the value is a dict mapping capacities (in Gbps) to probabilities for that link. Then, create a folder for your topology, exactly named, in the /hedge-te/data/inputs directory with the appropriate edges.txt and demand.txt files (see provided examples). Then, execute a Python script like this:

from NetworkTopology import *
from NetworkParser import *
from solver import *
from util import *

TOPOLOGY_FILENAME = "<PATH/TO/STOCHASTIC/TOPOLOGY/PKL>"
NETWORK_NAME = "<NAME OF NETWORK>"
DEMAND_FILENAME = "<PATH/TO/demand.txt>"

network,_ = get_max_and_min_networks(NETWORK_NAME, TOPOLOGY_NAME)
link_capacity_distributions = get_link_capacity_distributions_with_filename(topology_filename)
parse_demands(network, demand_filename, scale=demand_scale)
parse_tunnels(network)  # Default: 4 shortest paths

# Returns optimal bandwidth allocations along tunnels
results = solve_hedge(network, link_capacity_distributions)

Detailed Instructions

Hardware Experiments

Code and data for hardware experiments are available in the hardware-experiments/ folder. The only requirements are numpy and matplotlib.

• fiber_bend_wavelengths.ipynb generates Figures 4a and 4b. The data for this experiment can be found in the hardware-experiments/data/wdl/ folder.
• fiber_bend_modulation_formats.ipynb generates Figure 4c. The data for this experiment can be found in the hardware-experiments/data/mod_formats/ folder.
• prototype.ipynb generates Figure 14b and 14c. The data for this experiment can be found in the hardware-experiments/data/prototype/ folder.

Note: All transponder_data.csv files are in timestamp, channel, ber, fec, input_power format.

HEDGE-AGG Software

Code and data for the software evaluation of HEDGE-AGG are available in the hedge-agg/ folder. The requirements are pickle, matplotlib, and numpy.

• analysis.ipynb generates all subplots for Figure 6.
• snr_data.pkl contains real SNR data collected over 15 minute intervals for several months in an ISP WAN. This file contains a dictionary in which each key is an anonymized wavelength identifier and each value is a list of SNR values for that wavelength (in chronological order).
• snr_thresholds.pkl contains the SNR cutoff thresholds to sustain different data rates for 50 GHz spectral width and 32 GBaud baud rate. The file contains a list in which the value at index $i$ is the minimum SNR needed to sustain a data rate of $50 * i$ Gbps.

HEDGE-TE

Code and data for the evaluation of HEDGE-TE are available in the hedge-te/ folder. The requirements are gurobipy (Python API for the Gurobi Optimizer), pickle, matplotlib, and numpy. Our results for the B4 and ATT topologies can be visualized easily by running the analyze_results.ipynb notebook. For more advanced users, we delineate the following:

• Raw results from our evaluations on B4 and ATT are available in the hedge-te/data/results/<topology> folders (analyze_results.ipynb directly queries these). There are 10 results files for each topology since we ran the 1000 simulations for each of the 10 random permutations (of the capacity distribution-to-link mapping) for each topology.
• To run our extensive experiments from scratch yourself, you can run the run_experiments.py script, which has the usage: python run_experiments.py <path_to_stochastic_topology_file> <path_to_demand_file> <path_to_results_file>. The demand matrix and fixed topology (just showing the network structure, without link capacity distributions) files for B4 and ATT are available in the hedge-te/data/inputs/<topology> folders. <path_to_results_file> is completely up to your choosing.

Unfortunately, we cannot provide data for CloudWAN due to confidentiality requirements; this means we cannot provide the link capacity distributions for all topologies, since these are directly matched from CloudWAN data for link capacity fluctuations. To create your own stochastic topology (i.e., the first command-line argument for run_experiments.py script), you can create a pickle file containing a dict with format {<directed_edge>: {<capacity1>: <prob1>, <capacity2>, <prob2>,...}, ...} where the <directed_edge> key is a (str, str) tuple for a WAN link and the value is a dict mapping capacities (in Gbps) to probabilities for that link. For example, {('1', '2'): {1000: 0.99, 800: 0.009, 0: 0.001}, ('2', '1'): {1000: 0.995, 500: 0.005}}.

Demands and topologies for ATT and B4 are directly sourced from the TeaVaR repository.