Programmable Congestion Management SDK

This repository contains the source code for the PCM SDK.

PCM is a research and development SDK for implementing and evaluating programmable congestion control (CC) and load balancing (LB) algorithms. It provides CPU-based implementations of PCM algorithm interfaces and a runtime shell.

To compare supported algorithms under traffic patterns such as incast and core congestion, the project also integrates with UET htsim and ATLAHS/HTSIM toolchains, enabling both baseline network experiments and trace-driven studies.

SDK structure

SDK core:

• build.py: primary build entry point.
• hac/: code generation and helper tooling for algorithm integration.
• algorithms/: CC/LB algorithms.
• include/: PCM runtime.

Applications:

• apps/htsim/: UET htsim-based simulation app, scenarios, configs, and plotting scripts.
• apps/htsim_atlahs/: ATLAHS/GOAL-trace simulation app and related batch workflows.
• third-party/: external dependencies.

Testing PCM with applications

Dependencies

• C++23 compiler
• cmake >= 3.16
• numpy
• matplotlib
• pyyaml

Ultra Ethernet htsim application

1. Build UET htsim:

git submodule update --init --recursive
# The htsim patch exposes parts of the UE NIC/endpoint interface, 
# allowing CC/LB logic inside the UE stack to be redirected into PCM.
cp -r ./apps/htsim/uet-htsim-patch/* ./third-party/uet-htsim/htsim/sim/
cd ./third-party/uet-htsim/htsim/sim/
cmake -S . -B build
cmake --build build --parallel

2. Build xxHash

Load balancing algorithms rely on the xxHash library to provide fast PRNG:

cd ./third-party/xxHash && make

3. Build PCM runtime and htsim application

The PCM project can be built using build.py script which acts as a wrapper for CMake:

./build.py --relwithdebinfo

4. Test PCM congestion control algorithms with an incast traffic matrix

The following command can be used to run batch incast experiments for all supported CC algorithms:

export LD_LIBRARY_PATH=$(pwd)/build/lib/:$LD_LIBRARY_PATH
python3 ./apps/htsim/batch_simulations.py --conf=$(pwd)/apps/htsim/all_algos_incast.json --out=$(pwd)/results --plot

Upon successful execution, the ./results/ folder will contain congestion window evolution PDF plots for all algorithms listed in the all_algos_incast.json batch.

Other experiments (core congestion, LB algorithms) can be run using JSON batch configs in ./apps/htsim/.

ATLAHS+PCM: Testing PCM with GOAL application traces

PCM is integrated with the ATLAHS simulation toolchain, which extends UET htsim to support simulation of application GOAL traces.

See https://github.com/spcl/HTSIM for more information.

1. Build HTSIM_spcl

As with basic htsim, build spcl/HTSIM.

# make sure to pull submodules recursively
git submodule update --init --recursive
# apply HTSIM_spcl-patch to HTSIM_spcl first
cp -r ./apps/htsim_atlahs/HTSIM_spcl-patch/* ./third-party/HTSIM_spcl/htsim/sim/
# build htsim-atlahs
cd ./third-party/HTSIM_spcl/htsim/sim/
cmake -S . -B build
cmake --build build --parallel

2. Build xxHash

The procedure is similar to the UET htsim app.

3. Build ATLAHS+PCM application

python3 build.py --relwithdebinfo --build-htsim-atlahs

4. Test ATLAHS+PCM with incast trace

python3 ./apps/htsim_atlahs/batch_simulations_goal.py --conf ./apps/htsim_atlahs/pcm_goal_incast_16_nodes.json --out ./apps/htsim_atlahs/log_goal/16_nodes --plot

Useful build.py options

Help

python3 ./build.py --help

Cleanup build

python3 ./build.py --clean

Debug build with -O0

python3 build.py --debug

Profiling

python3 build.py --profiling --profiling-backend <perf,chrono,rdtsc>

Notes:

• The perf option is supported only on Linux. Run sudo sysctl -w kernel.perf_event_paranoid=0 to enable collection of perf data without root permissions.
• rdtsc is supported only on x86.