For CCGRID 2026 artifact reproduction instructions, see ARTIFACT.md.

Kumo: A Modular Serverless Security Simulator

Kumo is a modular, event-driven simulator for evaluating performance and security in multi-tenant serverless platforms.

This repository contains:

• The full Kumo simulator implementation

• The artifact accompanying the IEEE CCGRID 2026 paper:

  Kumo: A Security-Focused Serverless Cloud Simulator

Artifact DOI: https://doi.org/10.5281/zenodo.18635971

Overview

Kumo models:

• Function cold/warm starts
• Container lifecycle (idle timeout → cooling → eviction)
• Pluggable schedulers (Spread, Random, Helper, OpenWhisk, PASch, etc.)
• Workloads: Uniform, Poisson, Burst
• Multi-function tenants
• Worker heterogeneity
• Pre-warmed container pools
• Victim/attacker co-location attacks
• Config-driven experiments and parameter sweeps
• CSV-based metrics export
• Optional detailed tracing/logging

Kumo is designed to support reproducible experiments for cloud scheduling, co-location attacks, and serverless security research.

Project Structure

src/
  common/          Base types and utilities  
  config/          Config parser  
  core/            Engine, Events, Scenario, ExperimentRunner, TraceLogger  
  metrics/         Metric Collector  
  model/           Apps, functions, workers, containers  
  scheduler/       Pluggable schedulers + registry  
  workload/        Uniform, Poisson, Burst, Attack workloads  
  main_experiment.cpp  

configs/  
  Experiment configurations  

plot/  
  Python plotting scripts  

results/  
  Generated CSV logs  

tests/  
  Unit tests  

Build

Build the simulator:

make

This produces the binary: kumo_experiment

Run an experiment:

./kumo_experiment <config-file>

Core Capabilities

Schedulers

• spread — avoid tenant overlap (security-oriented)
• random — uniform random placement
• helper — warm-aware scheduling
• openwhisk — OpenWhisk-like placement
• openwhisk_warm — warm-enhanced OpenWhisk
• pasch — tenant packing strategy

Workloads

• Uniform — constant-rate batched arrivals
• Poisson — exponential inter-arrival times
• Burst — periodic high-rate spikes

All workloads support optional attacker traffic models.

Tenancy & Resource Modeling

• Multi-function tenants (functions_per_tenant=N)
• Worker heterogeneity (hetero.enabled=1)
• Idle timeouts for container eviction
• Pre-warmed container pools (prewarm.enabled=1)

Experiment Automation

Supports parameter sweeps over:

• schedulers
• random seeds
• idle timeouts

Exports metrics to CSV:

• cold/warm starts
• co-location counts
• per-tenant invocation stats
• worker stats

Optional detailed trace logging (trace.enabled=1).

Example Usage

Run a configuration file:

./kumo_experiment configs/example.cfg

Parameter sweep example:

./kumo_experiment configs/sweep_example.cfg

Artifact Reproduction (CCGRID 2026)

To reproduce the published CCGRID results:

make reproduce

See ARTIFACT.md for full reproduction instructions.

License

Released under the MIT License.