NetUber

Software-Defined Internet

NetUber is a research framework for simulating and orchestrating latency-sensitive web service workflows using a fleet of shared virtual routers (VRs) deployed on cloud spot instances. The framework acts as an evaluation platform for a Software-Defined Internet, enabling dynamic, bandwidth-aware, cost-optimized inter-cloud routing.

Architecture

graph TD
    subgraph "Control Plane (SDI Controller)"
        ORC[NetUber Orchestrator]
        P[Placement Engine]
        CP[Connectivity Provider]
    end

    subgraph "Messaging Bus (JMS / ActiveMQ)"
        MB[Topic: netuber.node.ID.cmds]
        ST[Topic: netuber.status.>]
    end

    subgraph "Data Plane (Cloud Nodes)"
        LA1[Local Agent - Node A]
        LA2[Local Agent - Node B]
        VR1[Docker: Virtual Router]
        VR2[Docker: Virtual Router]
    end

    ORC -->|Publish Cmds| MB
    MB -->|Consume| LA1
    MB -->|Consume| LA2
    LA1 -->|Exec| VR1
    LA2 -->|Exec| VR2
    LA1 -->|Telemetry| ST
    LA2 -->|Telemetry| ST
    ST -->|Feedback| ORC

Loading

Key Components

• Core Models: Node, Link, VirtualRouter, and Workflow.
• LatencyAwarePlacement: The engine responsible for calculating optimal deployment maps using bandwidth-aware heuristics and critical path latency verification.
• OverlayManager: Manages the lifecycle of the shared Virtual Router fleet and simulates regional price fluctuations.
• ConnectivityProvider: The Southbound interface using the SDI JSON Protocol over ActiveMQ.
• LocalAgent: A distributed agent running on cloud nodes that executes Docker-based VR deployments and generates BGP configurations.

Features

NetUber provides the following capabilities:

• DAG-Aware Orchestration: Workflows are modeled as Directed Acyclic Graphs (DAGs), enabling complex multi-stage pipelines with specific data volume constraints.
• Dynamic Spot Price Optimization: The orchestrator dynamically evaluates fluctuating spot prices across regions to minimize deployment costs.
• Bandwidth-Aware Routing: Evaluates outbound bandwidth requirements across the service graph to select capable deployment nodes.
• Resilient Recovery Loop: Provides automated recovery that re-places services during cloud spot instance preemption (runtime churn).
• Advanced Latency Verification: Uses critical path analysis to ensure compliance with service-level objectives (SLOs).
• Distributed Control Plane: Uses an event-driven model (ActiveMQ) for real-world orchestration of virtual router fleets.

Getting Started

Prerequisites

• Java 11+
• Maven 3.x
• ActiveMQ (Required for the Real SDI implementation. Run locally on tcp://localhost:61616)
• Docker (Required for actual routing container deployment)

Running the Real-World SDI Infrastructure

To run the distributed system (Agents + Controller):

./run_sdi.sh

This script compiles the project, launches Local Agents for three simulated regions, and starts the SDI Controller to orchestrate the fleet.

Running the Simulation Only

For a quick verification without distributed agents:

./run_simulation.sh

Documentation

For a detailed guide on deploying NetUber in a real cloud environment, see Tutorial.md.

Citing NetUber

If you use NetUber or the concept of a Software-Defined Internet in your research, please cite the following papers:

• Kathiravelu, P., Chiesa, M., Marcos, P., Canini, M. and Veiga, L., 2018, May. Moving bits with a fleet of shared virtual routers. In 2018 IFIP Networking Conference (IFIP Networking) and Workshops (pp. 1-9). IEEE.

• Kathiravelu, P., Van Roy, P., Veiga, L. and Benkhelifa, E., 2020, April. Latency-sensitive web service workflows: A case for a software-defined internet. In 2020 Seventh International Conference on Software Defined Systems (SDS) (pp. 115-122). IEEE.