https://github.com/spcl/evalnet
Official Implementation of "EvalNet: A Practical Toolchain for Generation and Analysis of Extreme-Scale Interconnects"
https://github.com/spcl/evalnet
interference multiplicity network-analysis network-topology networking shortest-paths
Last synced: 5 months ago
JSON representation
Official Implementation of "EvalNet: A Practical Toolchain for Generation and Analysis of Extreme-Scale Interconnects"
- Host: GitHub
- URL: https://github.com/spcl/evalnet
- Owner: spcl
- License: other
- Created: 2025-06-09T07:24:23.000Z (7 months ago)
- Default Branch: main
- Last Pushed: 2025-08-03T16:39:39.000Z (5 months ago)
- Last Synced: 2025-08-03T18:25:30.819Z (5 months ago)
- Topics: interference, multiplicity, network-analysis, network-topology, networking, shortest-paths
- Language: Python
- Homepage: http://arxiv.org/abs/2105.12663
- Size: 346 KB
- Stars: 3
- Watchers: 1
- Forks: 0
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# EvalNet
This is the official implementation of [EvalNet: A Practical Toolchain for Generation and Analysis of Extreme-Scale Interconnects](http://arxiv.org/abs/2105.12663).
EvalNet is a toolchain for scalable generation and analysis over 25 important network topologies, such as Slim Fly, PolarFly, and Orthogonal Fat Trees, with a strong focus on path diversity metrics.
EvalNet provides an extensive and fine-grained analysis of shortest and non-shortest paths, including their multiplicities, lengths, and interference.
## Setup Guide
In order to use this framework, you need to have a working installation of Python 3.9 or newer.
### Installing EvalNet
Before running the installation, make sure to activate your Python environment (if any) beforehand.
```
git clone https://github.com/spcl/EvalNet.git
cd EvalNet
pip install .
apt-get install libsqlite3-dev
cd analysis
edit Makefile
make
```
Our analysis code depends on libsqlite3, so please install its development files (header, static libraries) via your distributions packet manager.
Before compiling the source code in the [analysis](analysis) directory, please update the Python version (`PYTHON_VERSION`, default: 3.9) as well as the installation directory of your Python version (`PYTHON_DIR`) as well as your Python packages (`VIRTUALENV_DIR`) in [analysis/Makefile](analysis/Makefile).
If necessary, please update the C compiler (default: `gcc`) as well.
## Usage
We have integrated all currently supported functionality into a single tool.
```
python3 tool.py -h
usage: tool.py [-h] {generate,validate,clean,info,analyse,plot,show,ggplot,ggplot2} ...
positional arguments:
{generate,validate,clean,info,analyse,plot,show,ggplot,ggplot2}
type of operation
generate generates a topology
validate validates a topology
clean removes generated topologies
info saves information about selected topologies
analyse analysing tool
plot plotting tool
show Execute a SQL query and show the results.
ggplot Execute a SQL query and plot results using python-ggplot. Uses matplotlib, pandas and the ggplot package.
ggplot2 Execute a SQL query and plot results using R ggplot2. Requires sqldf and ggplot2 for R.
optional arguments:
-h, --help show this help message and exit
```
Specifically the tool supports the generation of new topologies (more in [topogen/README.md](topogen/README.md)), the analysis of already generated topologies as well as plotting the analyzed metrics (more in [analysis/README.md](analysis/README.md)).
## Supported Topologies
EvalNet supports the following topologies:
- n-dimensional hypercube
- k-ary n-dimensional torus
- k-ary n-flat (Flattened Butterfly)
- h-MLFM (Multi-Layer Full-Mesh)
- k-OFT (Two-Level k-Orthogonal Fat-Tree)
- Jellyfish (r-regular)
- regular HyperX
- Dragonfly (balanced)
- Fat-Tree (three-stage variant)
- Xpander
- 2x oversubscribed Fat-Tree
- Slim Fly
- Delorme
- Brown/Erdos-Renyi (ER) Graphs
- incremental extensions of Brown/ER graphs
- Bundlefly
- Kautz
- arrangement network
- extended generalized Fat Tree
- k-ary-n tree
- mesh
- Tofu
- Cascade Dragonfly
- Spectralfly
- Megafly
- Polarstar
The general command line interface for the topology generation can be found below:
```
usage: tool.py generate [-h]
{hypercube,torus,flatbutterfly,mlfm,oft,jellyfish,hyperx,dragonfly,fattree,xpander,slimfly,delorme,brown,brown_ext,bundlefly,kautz,arrnetwork,xgft,karyn,mesh,tofu,casdf,specfly,megafly,polarstar} ...
positional arguments:
{hypercube,torus,flatbutterfly,mlfm,oft,jellyfish,hyperx,dragonfly,fattree,xpander,slimfly,delorme,brown,brown_ext,bundlefly,kautz,arrnetwork,xgft,karyn,mesh,tofu,casdf,specfly,megafly,polarstar}
type of topology
hypercube generates a n-dimensional Hypercube topology
torus generates a k-ary n-Torus topology
flatbutterfly generates a k-ary n-flat (Flattened Butterfly)
mlfm generates a h-MLFM topology (Multi-Layer Full-Mesh)
oft generates a k-OFT topology (Two-Level Orthogonal Fat-Tree)
jellyfish generates a r-regular Jellyfish topology
hyperx generates a regular HyperX topology
dragonfly generates a Dragonfly topology
fattree generates a FatTree topology
xpander generates a Xpander topology
slimfly generates a SlimFly topology
delorme generates a Delorme topology
brown generates a Brown topology
brown_ext generates incremental expansions of a Brown topology
bundlefly generates bundlefly
kautz generates kautz
arrnetwork generates arrangement network
xgft generates extended generalized fat tree
karyn generates k-ary-n Tree
mesh generates Mesh k^n
tofu generates Tofu 6D Tofu
casdf generates Cascade Dragonfly (a=96,p=8,h=10) with g groups
specfly construct lps graphs (spectralfly)
megafly construct megafly
polarstar construct polarstar
optional arguments:
-h, --help show this help message and exit
```
We document the supported parameters for each topology in [topogen/README.md](topogen/README.md).
The output of the generated topologies is stored in the respective topology directory of the `data` directory.
## Supported Measures
EvalNet supports the following measures:
- shortest path length
- shortest path multiplicity
- count of edge disjoint paths
- path interference
The analyze part of the tool comes with the following command line interface:
```
usage: tool.py analyse [-h] {shortestpaths,disjointpaths,interference} ...
positional arguments:
{shortestpaths,disjointpaths,interference}
type of analysis
shortestpaths analyses shortest paths
disjointpaths analyses disjoint paths
interference analyses interference
optional arguments:
-h, --help show this help message and exit
```
We document the supported parameters for each measure in [analysis/README.md](analysis/README.md#measures).
Multiple topologies as well as the topologies classes, i.e. the number of nodes in the network, can be specified.
The output of the analysis is stored in `data/analysis/{analysis type}.db`.
## Supported Visualizations
EvalNet supports the following visualizations:
- histogram of shortest path length
- histogram of shortest path multiplicity
- histogram of count of edge disjoint paths
- 2-dimentional map of low connectivity pairs
- histogram of path interference
- detailed view of path interference
The respective tool part has the following command line interface:
```
usage: tool.py plot [-h] {shortestpaths,multiplicity,disjointpaths,interference,interferencedetail,lowconnectivity} ...
positional arguments:
{shortestpaths,multiplicity,disjointpaths,interference,interferencedetail,lowconnectivity}
type of plot
shortestpaths plots shortest paths
multiplicity plots shortest paths multiplicity
disjointpaths plots histogram of disjoint paths
interference plots histogram showing interference
interferencedetail plots histogram showing interference
lowconnectivity low connectivity plot
optional arguments:
-h, --help show this help message and exit
```
We document the supported parameters for each visualization in [analysis/README.md](analysis/README.md#visualizations).
## Citations
If you find this repository useful, please consider giving it a star! If you have any questions or feedback, don't hesitate to reach out and open an issue.
When using EvalNet in your work, please reference us with the citation provided below:
```bibtex
@misc{besta2025evalnet,
title = {{EvalNet: A Practical Toolchain for Generation and Analysis of Extreme-Scale Interconnects}},
author = {Besta, Maciej and Patrick Iff and Marcel Schneider and Nils Blach and Alessandro Maissen and Salvatore Di Girolamo and Jens Domke and Jascha Krattenmacher and Ankit Singla and Kartik Lakhotia and Laura Monroe and Fabrizio Petrini and Robert Gerstenberger and Hoefler, Torsten},
year = 2025,
month = Jun,
doi = {10.48550/arXiv.2105.12663},
url = {http://arxiv.org/abs/2105.12663},
eprinttype = {arXiv},
eprint = {2105.12663}
}
```