Diamond Hypercube Networking Project

This repo reconstructs and evaluates the Diamond–Hypercube (DQ) interconnection network alongside Mesh and 2D Torus topologies using Mininet + Open vSwitch.
The goal is to measure how topology impacts Flow Completion Time (FCT) under realistic background traffic (random-uniform and in-cast).

The Diamond Hypercube topology described in [1].

What's implemented

• Custom Mininet topologies:

  • Diamond Hypercube (DQ)
  • 2D Mesh
  • 2D Torus

• Shortest path routing algorithms:

  • X-Y routing for Mesh and Torus.
  • DQ routing:
    • Phase 1: route between groups by flipping group-label bits (hypercube-style).
    • Phase 2: route inside the diamond subgraph using precomputed next-hops (BFS-based).

• Traffic patterns (background load):

  • Random-uniform background flows: random source/destination pairs (seeded).
  • Incast (many-to-one) background flows: n random sources send to a single random destination (seeded).

• Metric: Flow Completion Time (FCT):

  • For each run, the system sends probe transfers between a fixed “farthest pair” in each topology, while background flows run concurrently. Probe completion time (from iperf3) is recorded as the FCT.
  • The median, 95th percentile and 99th percentile FCT is measured for the given topology and load level.

Requirements

• Linux enviromnemt
• Python 3.8+
• Mininet
• iperf3

How it works

1. Build topology in Mininet (hosts + OVS switches).
2. Install static routes by adding OpenFlow rules into OVS switches.
3. Start background traffic (random-uniform or in-cast).
4. Run repeated probes to measure FCT.
5. Append results to CSV and create graphs for median/95th percentile/99th percentile FCT.

Running experiments

1. Random uniform background traffic (single run):

   • Example (Mesh, r=14):

     • sudo python3 -m experiments.random_uniform_latency --topo mesh --r 14 --bg-num-flows 300 --bg-parallel-streams 1 --bg-duration 2000 --bg-warmup-sec 1 --num-probes 100 --probe-megabytes 10 --seed 1 --csv-out results_random_uniform.csv

2. In-cast background traffic (single run):

   • Exmaple (DQ, d=5):

     • sudo python3 -m experiments.incast_latency --topo dq --d 5 --bg-num-flows 300 --bg-parallel-streams 1 --bg-duration 2000 --bg-warmup-sec 1 --num-probes 100 --probe-megabytes 10 --seed 1 --csv-out results_incast.csv

3. Grid runs:

   • Run a predefined grid for random-uniform:

     • sudo python3 -m experiments.run_random_uniform_grid

   • Run a predefined grid for in-cast:

     • sudo python3 -m experiments.run_incast_grid

Plotting and Analysis

• A helper script computes median / p95 / p99 FCT vs background load and writes plots:

  • python3 analysis/analyze_fct_vs_bg.py --csv results_incast.csv --r 14 --d 5 --outdir plots

Citation

[1] A. Saif, O. Alhuniti, A. Abu Taleb, A. Odeh, and B. A. Mahafzah,
“Diamond hypercube interconnection network: topological structure and properties,”
The Journal of Supercomputing, vol. 81, article 901, 2025.
DOI: https://doi.org/10.1007/s11227-025-07379-4