Schedsim

A discrete event simulator for multicore real-time scheduling research. Schedsim models task execution on heterogeneous processors with support for:

• Real-time schedulers: EDF with CBS servers, GRUB/M-GRUB bandwidth reclamation, CASH
• Power management: DVFS (Power-Aware, FFA, CSF) and DPM (core sleep states)
• Energy analysis: Track power consumption and optimize for energy efficiency
• Heterogeneous platforms: ARM big.LITTLE and other multi-cluster architectures
• Multi-cluster allocation: First-fit, best-fit, worst-fit, capacity-aware, and adaptive allocators

Quick Start

# Build
nix build                    # or: cmake -S . -B build -G Ninja && cmake --build build

# Generate a task set
./result/bin/schedgen -n 10 -u 3.9 --umax 1.0 --period-mode range -d 100 -o scenario.json

# Run simulation with GRUB reclamation
./result/bin/schedsim -i scenario.json -p platforms/exynos5422LITTLE.json --reclaim grub -o trace.json

# Analyze results
./result/bin/schedview trace.json --energy --response-times

Installation

Using Nix (recommended)

nix build

Binaries are available at result/bin/{schedsim,schedgen,schedview,alloc}.

Using CMake

cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release
cmake --build build

Binaries are available at build/apps/{schedsim,schedgen,schedview,alloc}.

Command Reference

schedgen — Generate task sets

schedgen [OPTIONS]

Option	Description
-n, --tasks <n>	Number of tasks (required)
-u, --utilization <x>	Target total utilization (required, can exceed 1.0 for multicore)
--umin <x>	Min per-task utilization [0,1] (default: 0)
--umax <x>	Max per-task utilization [0,1] (default: 1)
-s, --success <x>	Success rate for deadline budget (0,1] (default: 1)
-c, --compression <x>	Compression ratio (min duration/WCET) [0,1] (default: 1)
--period-mode <mode>	harmonic (legacy fixed grid) or range (default: range)
--period-min <ms>	Min period in ms (default: 10, range mode)
--period-max <ms>	Max period in ms (default: 1000, range mode)
-d, --duration <s>	Simulation duration in seconds (range mode)
--seed <n>	Random seed
--batch <n>	Generate multiple scenarios
--dir <path>	Output directory for batch mode
-o, --output <file>	Output file (default: stdout)

schedsim — Run simulations

schedsim [OPTIONS]

Option	Description
-i, --input <file>	Scenario file, JSON (required)
-p, --platform <file>	Platform configuration, JSON (required)
--reclaim <policy>	Reclamation: none, grub, cash (default: none)
--dvfs <policy>	DVFS: none, power-aware, ffa, csf (default: none)
--dvfs-cooldown <ms>	DVFS cooldown in ms (default: 0)
--dpm <policy>	DPM: none, basic (default: none)
--dpm-cstate <n>	Target C-state (default: 1)
-d, --duration <s>	Simulation duration in seconds (default: auto)
--energy	Enable energy tracking
--context-switch	Enable context switch overhead
--format <fmt>	Output format: json, null (default: json)
--metrics	Print metrics to stderr
-o, --output <file>	Trace output file (default: stdout)

schedview — Analyze traces

schedview [OPTIONS] <trace-file>

Option	Description
<trace-file>	JSON trace file (positional)
-d, --directory <dir>	Process all *.json traces in directory
--deadline-misses	Show deadline miss details
--response-times	Show response time statistics
--energy	Show energy breakdown
--format <fmt>	Output format: text, csv, json (default: text)

alloc — Multi-cluster allocator testing

alloc [OPTIONS]

Option	Description
-i, --input <file>	Scenario file (required)
-p, --platform <file>	Platform configuration (required)
-a, --alloc <name>	Allocator (required): ff_big_first, ff_little_first, ff_cap, ff_cap_adaptive_linear, ff_cap_adaptive_poly, ff_lb, counting, mcts, first_fit, worst_fit, best_fit
-g, --granularity <mode>	per-cluster or per-core (default: per-cluster)
-A, --alloc-arg <k=v>	Allocator argument (repeatable)
--target <x>	u_target for LITTLE clusters
--reclaim <policy>	Reclamation: none, grub, cash
--dvfs <policy>	DVFS: none, power-aware, ffa, csf, ffa-timer, csf-timer

Platforms

Platform definitions in platforms/ describe hardware configurations including:

• Processor types and clusters (e.g., big.LITTLE)
• Available frequencies and voltage levels
• Power models for energy estimation

See platforms/exynos5422.json for a complete ARM big.LITTLE example.

Architecture

Schedsim uses a three-library architecture:

Library	Path	Role
core	schedsim/core/	Simulation engine, hardware model, events
algo	schedsim/algo/	Schedulers, policies, allocators
io	schedsim/io/	JSON loading, trace output, scenario generation

See docs/architecture.md for details.

Testing

# Run all unit tests
cd build && ctest --output-on-failure

# Run GRUB verification suite (1000+ scenarios)
python tests/verify_grub.py --schedsim build/apps/schedsim