HDE2E Demo and Analysis Guide

This document describes how to run the High Density End-2-End (HDE2E) demo benchmark and analyze the results.

Prerequisites

uv package manager installed
Network proxy configured (if behind corporate firewall)
Docker installed and running
sudo access for hardware-level operations

Quick Start

1. Run the HDE2E Demo

# If NOT behind a proxy:
sudo $(which uv) run main.py

# If behind a proxy, set environment variables:
sudo http_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT https_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT $(which uv) run main.py

Replace YOUR_PROXY_HOST and YOUR_PROXY_PORT with your proxy settings (e.g., proxy.company.com:8080).

What this does:

Collects system information
Applies PQOS cache and memory bandwidth optimizations (if enabled)
Launches CODESYS control and IO PLC instances
Sets up Docker networking and port forwarding
Runs the HDE2E latency and jitter benchmark

Expected output:

System info saved to outputs/YYYY-MM-DD/HH-MM-SS/sysinfo.json
Docker containers running:
- Control_PLC_01 on cores 3,5
- Control_PLC_02 on cores 7,9
- IO_PLC_01 and IO_PLC_02 on remote system (10.34.106.119)
Port forwarding configured:
- Control web UI: ports 30080, 30081
- IO web UI: ports 30090, 30091
- CODESYS gateway: ports 11741, 11742

2. Retrieve Results from IO System

The benchmark generates result files on the remote IO system. Copy them to your local machine:

# Create results directory
mkdir -p ~/Results_testing

# Copy result files from IO system (10.34.106.119)
# Results are stored in: ~/dockerMount/IO_PLC_01/data/codesyscontrol/PlcLogic/Results/
scp intel@10.34.106.119:~/dockerMount/IO_PLC_01/data/codesyscontrol/PlcLogic/Results/*.csv ~/Results_testing/

Note: You may need to use the same SSH password from conf/config.yaml or set up key-based authentication.

3. Analyze Results

# If NOT behind a proxy:
sudo $(which uv) run hde2e-analyze /path/to/results [options]

# If behind a proxy:
sudo http_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT https_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT $(which uv) run hde2e-analyze /path/to/results [options]

Options:

-v, --verbose - Print detailed statistics
-d, --debug - Print debug information
--save - Save plots as PNG files
--show - Display plots interactively (use only with single instance for testing)
--rows N - Read only first N rows of input data
--version - Show version

Example with output:

sudo http_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT https_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT $(which uv) run hde2e-analyze /home/intel/Results_testing -v --save

Output files generated:

stat_Latency.csv - Latency statistics
stat_Jitter.csv - Jitter statistics
box_Latency.png - Box plot for latency
box_Jitter.png - Box plot for jitter
bar_Latency.png - Bar chart for latency statistics
bar_Jitter.png - Bar chart for jitter statistics
scatter_*.png - Scatter plots for each latency component and jitter

Configuration

Demo Mode Settings

Edit conf/config.yaml to customize the demo:

demo:
  demo_mode: true              # Enable demo mode
  t_core: [9,11]              # Target cores
  io_system:
    ip: 10.34.106.119         # Remote IO system IP
    nic: enp3s0               # Network interface
    ssh_user: intel           # SSH username
    ssh_password: "..."       # SSH password
    ssh_port: 22              # SSH port
    t_cpus: "1,3"             # IO system target CPUs
  control_system:
    ip: localhost             # Control system IP
    nic: eno12399np0          # Network interface
    t_cpus: "3,5,7,9"         # Control CPUs

PQOS Optimizations

PQOS settings are automatically applied when running the demo. Configure them in conf/config.yaml:

pqos:
  interface: "os"              # Use Linux resctrl interface
  reset_before_apply: true     # Reset before applying
  
  classes:
    - id: 1
      description: "real-time workload"
      l3_mask: "0x00ff"        # L3 cache mask (8 ways)
      l2_mask: "0x00ff"        # L2 cache mask (8 ways)
      mba: 100                 # Memory bandwidth 100%
      pids: []                 # PIDs (empty for demo)
      cores: []                # Cores (empty for demo)
      
    - id: 0
      description: "background worker"
      l3_mask: "0x7f00"        # L3 cache mask (7 ways)
      l2_mask: "0xff00"        # L2 cache mask (8 ways)
      mba: 10                  # Memory bandwidth 10%
      cores: [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
      pids: [115, 118]

Complete Workflow Example

# Set proxy variables (if behind a proxy)
export http_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT
export https_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT

# 1. Run the demo (takes ~5-10 minutes)
sudo $(which uv) run main.py

# 2. The benchmark completes - result files are generated on IO system

# 3. Copy results from remote IO system to local machine
mkdir -p ~/Results_testing
scp intel@10.34.106.119:~/dockerMount/IO_PLC_01/data/codesyscontrol/PlcLogic/Results/*.csv ~/Results_testing/

# 4. Verify results were copied
ls -lh ~/Results_testing/Codesys-*.csv

# 5. Analyze results with statistics and plots
sudo $(which uv) run hde2e-analyze ~/Results_testing -v --save

# 6. View generated files
ls -lh ~/Results_testing/*.{csv,png}

Cleaning Up

Stop Running Containers

sudo docker stop Control_PLC_01 Control_PLC_02
sudo docker rm Control_PLC_01 Control_PLC_02

Kill Port Forwarding

sudo pkill -f socat

Full Cleanup (if containers are stuck)

sudo systemctl restart docker

Troubleshooting

SSH / SCP Issues

Error: Permission denied (publickey,password)

Ensure you have the correct SSH password from conf/config.yaml

Try with explicit password prompt:

scp -P 22 intel@10.34.106.119:/home/intel/dockerMount/IO_PLC_01/data/codesyscontrol/PlcLogic/Results*.csv ~/Results_testing/

If using key-based auth, ensure your public key is on the IO system

Error: No such file or directory

Verify the results directory path on the IO system
Connect to IO system to check: ssh intel@10.34.106.119 ls -la /home/intel/dockerMount/IO_PLC_01/data/codesyscontrol/PlcLogic/Results
Results may be in a different location; check the benchmark output logs

Slow SCP Transfer

Results files can be large (20MB+)
Use compression: scp -C intel@10.34.106.119:/path/to/results/*.csv ~/Results_testing/

Network Issues

Error: Failed to connect to github.com

Ensure proxy is configured: git config --global http.proxy http://YOUR_PROXY_HOST:YOUR_PROXY_PORT
Or set environment variables: export http_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT

Error: Failed to fetch from pypi.org

Run with proxy environment variables:

sudo http_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT https_proxy=http://YOUR_PROXY_HOST:YOUR_PROXY_PORT $(which uv) ...

Or configure permanently in git: git config --global http.proxy http://YOUR_PROXY_HOST:YOUR_PROXY_PORT

SSH Authentication

Error: Authentication failed for IO system

Check SSH credentials in conf/config.yaml
Verify network connectivity to 10.34.106.119 (IO system IP Address)
Ensure SSH password is correct

PQOS Issues

Error: pqos tool is not installed

The demo will skip PQOS if not available
Install with: sudo apt-get install intel-cmt-cat

Error: ROOT ACCESS REQUIRED

All operations require sudo for hardware access

Docker Issues

Error: Container stuck in exited state

Restart Docker: sudo systemctl restart docker
Force remove: sudo docker rm -f <container_id>

Output Format

Statistics Output

The analysis script produces statistics in multiple formats:

Console Output (verbose mode):

#### T2-T1 ############################
       mean   std   min   max
IO_01 355.4 144.5 195.0 754.0

CSV Files:

Statistics at percentiles: 0.9, 0.99, 0.999, 0.9999, 0.99999
One row per instance
Columns: mean, std, min, max, and percentiles

Latency Components

T2-T1: IO to Control transmission time
T3-T2: Control processing time
T4-T3: Control to IO transmission time
T4-T1: Total end-to-end latency
Jitter: PubSub cycle time variation

Performance Metrics

The HDE2E benchmark measures:

Latency: Time for data to traverse from IO → Control → IO (in microseconds)
Jitter: Variation in PubSub task cycle time (in microseconds)

With PQOS optimizations enabled:

Real-time workload (Class 1) gets priority cache access
Background workers (Class 0) are isolated with limited bandwidth
Results in lower latency and reduced jitter variance

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

HDE2E Demo and Analysis Guide

Prerequisites

Quick Start

1. Run the HDE2E Demo

2. Retrieve Results from IO System

3. Analyze Results

Configuration

Demo Mode Settings

PQOS Optimizations

Complete Workflow Example

Cleaning Up

Stop Running Containers

Kill Port Forwarding

Full Cleanup (if containers are stuck)

Troubleshooting

SSH / SCP Issues

Network Issues

SSH Authentication

PQOS Issues

Docker Issues

Output Format

Statistics Output

Latency Components

Performance Metrics

References

FilesExpand file tree

DEMO_AND_ANALYSIS.md

Latest commit

History

DEMO_AND_ANALYSIS.md

File metadata and controls

HDE2E Demo and Analysis Guide

Prerequisites

Quick Start

1. Run the HDE2E Demo

2. Retrieve Results from IO System

3. Analyze Results

Configuration

Demo Mode Settings

PQOS Optimizations

Complete Workflow Example

Cleaning Up

Stop Running Containers

Kill Port Forwarding

Full Cleanup (if containers are stuck)

Troubleshooting

SSH / SCP Issues

Network Issues

SSH Authentication

PQOS Issues

Docker Issues

Output Format

Statistics Output

Latency Components

Performance Metrics

References