Skip to content

compu-TEE/SRCC-implementation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SRCC Implementation

Implementation of SRCC (Sub-RTT Congestion Control) for lossless RDMA datacenter networks.

This project reproduces and evaluates the congestion control mechanism proposed in the paper:

SRCC: Sub-RTT Congestion Control for Lossless Datacenter Networks (IEEE Transactions on Industrial Informatics, 2025)


Overview

Traditional RDMA congestion control protocols such as DCQCN, HPCC, Swift, and PCN rely on RTT-scale feedback loops. Under bursty traffic conditions, congestion often develops faster than these protocols can react, leading to:

  • Queue buildup
  • PFC triggering
  • Head-of-Line (HoL) blocking
  • Throughput degradation

SRCC introduces a sub-RTT feedback mechanism that allows switches to directly notify senders about congestion before a full RTT elapses.

This repository contains an implementation of SRCC along with experiments comparing it against:

  • DCQCN
  • HPCC
  • Swift
  • PCN

Features

  • Dynamic congestion threshold calculation
  • Congested flow identification
  • Congestion Notification Message (CNM) generation
  • Sender-side rate adjustment
  • Queue monitoring
  • Performance evaluation framework
  • Comparison with existing RDMA congestion control schemes

Implementation Details

The implementation includes:

  • SRCC congestion detection logic
  • Congestion threshold computation
  • Congestion notification generation
  • Sender-side rate control
  • Metrics collection and visualization

Key metrics evaluated:

  • Queue Length
  • Pause Rate
  • Throughput

Results

Queue Length

SRCC reacts significantly faster to congestion, resulting in reduced queue buildup.

Queue Length


Pause Rate

SRCC minimizes pause events by preventing prolonged congestion.

Pause Rate


Throughput

SRCC maintains high throughput even under bursty traffic conditions.

Throughput


Combined Comparison

Combined Metrics


Experimental Observations

From the reproduced experiments:

  • SRCC exhibits the lowest queue occupancy.
  • Pause events are significantly reduced.
  • Throughput remains close to line rate.
  • Recovery after congestion is substantially faster.
  • Congestion persists for a shorter duration compared to DCQCN, HPCC, Swift, and PCN.

Repository Structure

SRCC-implementation/
│
├── src/
├── configs/
├── scripts/
├── assets/
│   ├── combined_metrics_graph.png
│   ├── queue_length_graph.png
│   ├── pause_rate_graph.png
│   └── throughput_graph.png
│
├── README.md
└── ...

Reference

Hu, Jinbin et al.

SRCC: Sub-RTT Congestion Control for Lossless Datacenter Networks

IEEE Transactions on Industrial Informatics, 2025.

Paper: https://ieeexplore.ieee.org/document/10813045

DOI: https://doi.org/10.1109/TII.2024.3495759

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages