SwitchSTM: Enabling Transaction Switch in Loop-based Software Transactional Memory Applications via Coroutines

SwitchSTM enables transaction switch built on the top of TinySTM. SwitchSTM aims to address the furtile stall issue with lightweight coroutine switching.

Table of Contents

• Requirements
• Installation
• Running Simulations
  • Command Syntax
  • Available STM Labels
  • Examples
• Switch Strategies
  • Base Strategies
  • Additional Configurations
  • Recommended Flags
• Analysis and Plotting
  • Plotting Comparison Results (plot.py)
  • Analyzing Benchmark Results (parse_stm.py)
  • Profiling Execution Time Breakdown
  • Generating and Plotting CSV Data
• Development Guide
  • Project Structure
  • Makefile Configuration
  • Adding a New Strategy

Requirements

• GNU gcc/g++ version >= 9.4.0
• make
• libaco

Installation

Clone this repository with all submodules:

git clone --recurse-submodules https://github.com/NTUDDSNLab/SWITCH_STM.git

Running Simulations

The primary entry point is simulation.py, which automates the build and execution process.

Command Syntax

python3 simulation.py -simulation_times <N> -threads_list "<threads>" -log_path <path> <STM_labels>

Available STM Labels

Baselines:

• -suicide
• -polka
• -shrink
• -ats

SWITCH_STM Variants:

• switch_rnd: Random switch
• switch_seq: Sequential switch
• switch_laf: Less Abort First
• Suffixes: _CI (Contention Intensity), _TP (Time Profiling)
  • e.g., -switch_rnd_CI, -switch_seq_TP, -switch_laf_CI_TP

Examples

1. Basic Run: Run suicide, polka, and random switch strategies with 1 simulated run.

python3 simulation.py -suicide -polka -switch_rnd

2. Comprehensive Simulation: Run 10 iterations, identifying thread scaling (1, 2, 4, 8, 16), with profiling enabled.

python3 simulation.py -simulation_times 10 -thread "1 2 4 8 16" -suicide -polka -shrink -switch_rnd_TP -switch_seq_TP

Switch Strategies

Base Strategies

1. switch_rnd ("Random"): The switcher randomly selects a coroutine to switch to.
2. switch_seq ("Sequential"): Coroutines are selected in specific order based on unique identifiers.
3. switch_laf ("Less Abort First"): Selects the coroutine with the lowest current abort count.

Additional Configurations

Append these suffixes to the base strategy (e.g., -switch_rnd_CI_TP).

1. _CI (Contention Intensity):

   • Calculates current conflict level.
   • Switching only occurs if conflicts exceed a threshold.
   • Low contention -> direct restart (avoids switch overhead).

2. _TP (Time Profile):

   • Enables detailed time profiling (commit, abort, wait, switch time).
   • Note: If combining with CI, place TP after CI (e.g., _CI_TP).

Recommended Flags

When compiling/running SWITCH_STM manually or verifying configuration:

• CM=CM_SUICIDE
• IRREVOCABLE_ENABLED
• EPOCH_GC

Analysis and Plotting

Plotting Comparison Results (plot.py)

Generates plots from raw log data.

1. Prepare Data: Create a subfolder inside plot/ and place your raw .stm log files there.

Profiling Execution Time Breakdown

When SWITCH_STM_TIME_PROFILE is enabled (e.g., -switch_rnd_TP):

1. Parse Logs: Use tables/parse_profile.py to generate a CSV summary of breakdown metrics (Commit, Abort, Wait, Switch, Other).

python3 tables/parse_profile.py --log_dir ./log --output profile_results.csv

Generating and Plotting CSV Data

1. Generate CSV from Logs: Convert all .stm logs in a directory to a single CSV file.

python3 tables/generate_csv.py [log_dir] --output stm_results.csv

2. Plot from CSV: Use plot/plot_csv.py to plot data from the CSV.

# Plot execution time (default)
python3 plot/plot_csv.py stm_results.csv suicide switch_rnd

# Plot abort ratio
python3 plot/plot_csv.py stm_results.csv suicide switch_rnd -m abort

Development Guide

Project Structure

• include/param.h: Parameters for Switch_STM.
• include/switch_table.h: Header for switch table management.
• src/switch_table.c: Functions for creating/deleting switch tables and switching coroutines.
• src/switcher.h: Algorithm flow, initialization, and strategy logic.

Makefile Configuration

The simulation.py script automatically handles these flags:

• SWITCH_STM: Enables SWITCH_STM functionality.
• CONTENTION_INTENSITY: Enables CI detection.
• SWITCH_STM_TIME_PROFILE: Enables detailed timing collectors.

Adding a New Strategy

1. Modify src/switcher.h:
   • In switcher_decide(), add a new case for your strategy.
   • Return the integer decision for the chosen coroutine.
2. Update simulation_function.py:
   • Update simulate_switch_stm() to handle the new flag and map it to the code added in switcher.h.