Introducing ENTE-TCP: Smart TCP Congestion Control for Modern Networks

A Simplified Approach to Intelligent Congestion Control

ENTE-TCP is a streamlined version of NDM-TCP (Neural Differential Manifolds TCP), designed to bring the core innovation of entropy-based congestion detection to a wider audience without the complexity of machine learning.

What makes ENTE-TCP different from NDM-TCP?

NDM-TCP pioneered the use of neural networks combined with Shannon Entropy for TCP congestion control, achieving excellent performance through adaptive learning.

ENTE-TCP takes the proven entropy calculation from NDM-TCP and replaces the neural network layers with straightforward, transparent logic. The result is an algorithm that's:

• Easy to understand: Clear if-else decisions based on entropy thresholds
• Lightweight: No neural network overhead
• Transparent: Every decision is explainable
• Effective: Retains the core noise vs. congestion detection capability

Think of it as "NDM-TCP without the ML" — keeping the brilliant insight about entropy, removing the neural network complexity.

The Problem with Traditional TCP

Modern networks face a critical challenge: not all packet loss means congestion.

Traditional TCP algorithms like Reno and CUBIC were designed in an era of reliable wired networks. They operate on a simple assumption: when packets are lost or RTT increases, the network must be congested. So they aggressively reduce the sending rate.

But this assumption breaks down in today's wireless-dominated world:

• WiFi interference causes random packet drops
• Mobile handoffs between cell towers create temporary disruptions
• Signal fluctuations in wireless networks introduce RTT jitter
• Channel switching in crowded wireless environments causes instability

When traditional TCP encounters these issues, it mistakes random noise for congestion and unnecessarily throttles throughput. The result? Poor performance on mobile networks, WiFi, and satellite connections.

Enter ENTE-TCP: Intelligence Through Entropy

ENTE-TCP (Entropy-Enhanced TCP) takes a fundamentally different approach. Instead of treating all packet loss equally, it asks: "Is this random noise, or real congestion?"

The answer lies in Shannon Entropy — a mathematical measure from information theory that quantifies randomness.

How Entropy Reveals the Truth

Think of RTT measurements as a signal:

Scenario 1: Random Noise (WiFi Interference)

RTT pattern: 20ms → 45ms → 18ms → 52ms → 23ms → 48ms
Analysis: Values jump randomly, high unpredictability
Entropy: HIGH (≈0.85)
Interpretation: This is noise, not congestion
Action: Stay aggressive, maintain throughput

Scenario 2: Real Congestion (Queue Buildup)

RTT pattern: 20ms → 25ms → 30ms → 35ms → 40ms → 45ms
Analysis: Steady increase, predictable pattern
Entropy: LOW (≈0.35)
Interpretation: Queues are building, real congestion
Action: Back off conservatively

The key insight: random patterns indicate noise, consistent patterns indicate congestion.

The Algorithm in Action

ENTE-TCP continuously monitors the last 16 RTT samples and calculates their entropy every 8 packets:

1. Entropy Classification

• High Entropy (>0.7): Random variation → Likely wireless noise
• Low Entropy (<0.4): Consistent pattern → Likely real congestion
• Medium Entropy (0.4-0.7): Unclear → Use standard TCP behavior

2. Adaptive Response

When noise is detected (high entropy):

• Grow congestion window 1.5× faster during congestion avoidance
• Reduce congestion window to only 2/3 on packet loss (instead of 1/2)
• Rationale: Don't overreact to random events

When congestion is detected (low entropy):

• Grow congestion window 0.5× slower during congestion avoidance
• Reduce congestion window to standard 1/2 on packet loss
• Rationale: Respect genuine network limits

3. Continuous Adaptation

ENTE-TCP doesn't make one-time decisions. It continuously updates its understanding of network conditions, adapting in real-time as the environment changes.

Real-World Benefits

For Mobile Users

• Faster downloads on cellular networks despite signal fluctuations
• Smoother video streaming that doesn't pause during handoffs
• Better video calls that maintain quality through interference

For WiFi Networks

• Higher throughput in congested WiFi environments
• Reduced sensitivity to neighboring network interference
• Improved performance in coffee shops, airports, public spaces

For Satellite/Long-Distance Links

• Better handling of variable latency
• Faster recovery from temporary disruptions
• Maintained throughput despite atmospheric conditions

Technical Advantages

Simplicity: Pure mathematical approach using information theory, no machine learning black boxes

Efficiency: Lightweight computation (entropy calculation on 16 samples every 8 packets)

Compatibility: Works with existing TCP infrastructure, no protocol changes needed

Transparency: All decisions based on clear entropy thresholds, fully explainable

Memory Efficient: Only ~80 bytes per connection, fits standard kernel structures

How It Compares

(based on tests ,not tested on real hardware- tested on linux using tc)

ENTE-TCP vs NDM-TCP:

• Same core insight: Both use Shannon Entropy to distinguish noise from congestion
• Different approach: NDM-TCP uses neural networks for decisions; ENTE-TCP uses simple thresholds
• Performance: NDM-TCP may achieve slightly better adaptation through learning; ENTE-TCP is more predictable and easier to tune
• Complexity: NDM-TCP requires understanding of neural networks; ENTE-TCP is straightforward logic
• Use case: Choose NDM-TCP for cutting-edge performance with ML; choose ENTE-TCP for transparent, maintainable performance

ENTE-TCP shines in mixed environments where wireless and wired paths coexist, and in scenarios where algorithm transparency and maintainability are priorities.

The Science Behind It

ENTE-TCP leverages Shannon Entropy, defined as:

H = -Σ p(x) × log₂(p(x))

Where:

• H = entropy (measure of randomness)
• p(x) = probability of each RTT value
• Higher H = more random = likely noise
• Lower H = more predictable = likely congestion

This isn't guesswork — it's a fundamental principle of information theory that has been proven for over 75 years since Claude Shannon's groundbreaking work in 1948.

Getting Started

ENTE-TCP is implemented as a Linux kernel module:

# Build the module
make

# Load and test
make test

# Check status
make status

Once loaded, set it as your default congestion control:

sudo sysctl -w net.ipv4.tcp_congestion_control=ente_tcp

Open Source and Available Now

ENTE-TCP is open source under GPL v2, the same license as the Linux kernel.

Repository: https://github.com/hejhdiss/lkm-ente-tcp

Whether you're a researcher exploring congestion control, a network engineer optimizing performance, or a developer working on mobile applications, ENTE-TCP offers a fresh approach to an old problem.

The Bottom Line

In a world where wireless networks are ubiquitous and network conditions constantly change, we need congestion control that's smart enough to tell the difference between a noisy WiFi signal and an overloaded router.

ENTE-TCP does exactly that — using the elegant mathematics of entropy to make TCP faster, smarter, and more resilient in the real world.

Built on the proven foundation of NDM-TCP's entropy-based approach, ENTE-TCP offers a transparent, maintainable alternative that brings intelligent congestion control to those who prefer clarity over complexity.

Read Time: ~3 minutes

Learn More: https://github.com/hejhdiss/lkm-ente-tcp

Original NDM-TCP: https://github.com/hejhdiss/lkm-ndm-tcp

License: GPL v2