iDynamics: A Configurable Emulation Framework for evaluating Microservice Scheduling Policies under Controllable Cloud–Edge Dynamics

iDynamics is a configurable emulation framework for evaluating microservice scheduling policies on Kubernetes-based cloud–edge clusters under controllable dynamics (call-graph, traffic, and network conditions).

The repository hosts the research prototype used in our iDynamics papers and provides a basis for implementing and testing new scheduling policies against realistic but repeatable conditions.

---

What is “controllable dynamics”?

In iDynamics, controllable dynamics means you can systematically:

• reconstruct and vary microservice call-graph topologies,
• stress specific service chains with different traffic mixes, and
• inject realistic cross-node latency and bandwidth profiles,

while observing the resulting SLA metrics (e.g., tail latency and throughput).

This allows you to ask questions such as:

• How does a given scheduler behave under increasingly skewed traffic?
• Under which delay / bandwidth patterns does it start to violate latency SLOs?
• Which microservice pairs become bottlenecks as the request mix evolves?

---

Key capabilities

iDynamics is designed to:

• Run real microservices on a Kubernetes cluster (e.g., DeathStarBench Social Network) instead of pure simulation.
• Observe dynamic call-graphs and traffic via a service mesh (Istio) and build weighted dependency graphs between upstream–downstream microservice pairs.
• Emulate heterogeneous cross-node latency and bandwidth using Linux traffic control prmitives (classful qdisc + u32 filters) while preserving non-experimental traffic.
• Measure injected dynamics with lightweight, distributed agents (Kubernetes DaemonSets) for delay and bandwidth.
• Plug in arbitrary scheduling policies through an abstract policy interface and a utility module that exposes nodes, pods, and metrics in a scheduler-friendly way.
• Evaluate policies under repeatable scenarios, including dynamic workloads, changing call-graphs, and evolving network conditions.

---

Architecture at a glance

At a high level, iDynamics consists of three main components:

1. Graph Dynamics Analyzer

   • UM–DM Traffic Profiler
   • Call-Graph Builder
     Reconstructs call-graphs from service-mesh telemetry and computes stress (bi-directional traffic) between microservice pairs over a time window.

2. Networking Dynamics Manager

   • Emulator
     Injects configurable delay and bandwidth profiles per node pair using Linux tc (HTB + filters).
   • Measurer
     Distributed agents (DaemonSets) + a central collector to measure and validate cross-node delays and bandwidths.

3. Scheduling Policy Extender

   • Policy Customization Interface
     An abstract AbstractSchedulingPolicy class defining hooks for single-pod and batch scheduling, plus metric updates.
   • Utility Function Module
     Helpers to obtain node/pod state, SLA targets, and metrics from Prometheus / service mesh, and to push scheduling decisions.

Example policies implemented in the papers include:

• Policy 1 – Call-graph–Aware: co-locates heavily communicating microservices or places them on low-latency nodes.
• Policy 4 – Hybrid-dynamics–Aware: minimizes a service–node mapping cost that combines traffic stress and cross-node delay.

---

Typical use cases

You can use iDynamics to:

• Prototype and debug new microservice scheduling policies (heuristic or ML-based) before touching production.
• Study SLA compliance under controlled workload and infrastructure dynamics.
• Compare network-aware vs. network-agnostic scheduling strategies.
• Explore trade-offs between communication efficiency and resource balance in cloud–edge clusters.

---

Prerequisites

To use iDynamics as in the papers, you will typically need:

• A Kubernetes cluster (tested with 5–15 worker nodes in the papers) with:
  • Linux nodes and tc (traffic control) available.
  • A CNI plugin such as Calico.
  • Istio service mesh enabled for your microservices.
• Python 3.10+ with pip.
• Permission to deploy DaemonSets and custom controllers/plugins on the cluster.
• (Optional, for reproducing case studies)
  • [DeathStarBench] Social Network microservice application and a workload generator (e.g., wrk2).

---

Citation:

@misc{chen2025idynamicsnovelframeworkevaluating, title={iDynamics: A Configurable Emulation Framework for evaluating Microservice Scheduling Policies under Controllable Cloud–Edge Dynamics}, author={Ming Chen and Muhammed Tawfiqul Islam and Maria Rodriguez Read and Rajkumar Buyya}, year={2025}, eprint={2503.16029}, archivePrefix={arXiv}, primaryClass={cs.DC}, url={https://arxiv.org/abs/2503.16029}, }