Ginpex

Overview

Ginpex (Goal-oriented INfrastructure Performance EXperiments) is being developed for

• Running experiments that induce certain load patterns on machines
  • Use Ginpex as load generator
  • Distribute load to multiple machines
  • Convenient experiment specification with the experiment model editor
  • Add machines dynamically to running experiments
  • Automate experiments
  • Integrate your own experiment logic into Ginpex experiment models
• Automatically deriving performance-relevant infrastructure parameters
  • Ginpex ships with predefined experiments
  • Use detected parameters for performance prediction, for example as input for the PCM

Infrastructure parameters in focus include

• OS scheduling parameters (timeslice length, load balancing policies)
• Resource parameters (HDD, network)
• Virtualization parameters (Virtualization overhead, ...)

Additional features include

• A UI editor to comfortably specify experiment specifications
• Java code generation based on specified experiment specifications for efficient experiment execution on any platform where Java is available
• Checkpoint functionality (snapshots) for long-running experiments

Approach

The main application scenario of Ginpex is embedded in model-based software performance prediction.

Architecture

The architecture is based on the Performance Cockpit approach [1].

Metamodel

Metamodel in Detail

Extensions

Available Ginpex extensions

Installation

Ginpex installation instructions

How to use Ginpex

Have a look at the Ginpex Cookbook for help on using Ginpex.

Predefined experiments

sdqinternal:Ginpex/Predefined Experiments

Including Virtualization Overheads in Performance Prediction

Contact

Ginpex Homepage

For questions, contact Michael Hauck.