Temporal Predictive Replica Placement in FogStore - A Simulation Framework

This simulation is a fork from the original prp-simulation extending it by the temporal prediction, thus being able to simulate spatio-temporal prediction models.

Structure:

• geolife-data: location of the raw Geolife Data
• geolife-data-transformed: populated by the me.mbe.prp.TransformGeolife main method in the test module
• src code
  • main: code of the framework, algorithms, etc.
    • algorithms
    • base: helper methods
    • core: framework
    • data
    • metrics
    • network
    • nodes
  • test: code for the evaluation
• stats-out: evaluation results saved here
• analysis: python files for analysis and plotting of the results

---

Setup:

1. Clone repository
2. Setup gradle
3. Copy the Geolife Data into the geolife-data folder (copy the whole Data folder).
4. Run the me.mbe.prp.TransformGeolife main method in the test module
5. Run the evaluations in me.mbe.prp.geolife.Evaluation
6. Results can be found in the stats-out directory (The results folder already contains the results files, in order to create new ones, delete the existing ones.)

---

Algorithms:

Spatio-temporal algorithms (our proposed models):

• Baseline
  • me.mbe.algorithms.Alg000: Store data on all nodes at all times.
  • me.mbe.algorithms.Alg001: Store data only on closest node when application active.
  • me.mbe.algorithms.nextnodepred.Alg004: Variable Order Markov Model.
  • me.mbe.algorithms.nextnodepred.Alg012: Fusion Multi Order Markov Model.
• T-VOMM
  • me.mbe.algorithms.nextnodepred.AlgT04: Temporal Variable Order Markov Model: Not mentioned in the thesis
• T-FOMM
  • me.mbe.algorithms.nextnodepred.AlgT012: Temporal Fusion Multi Order Markov Model with multiple temporal models: Percentiles (PCTL), Temporal Discretization (TD) and Holt Winter’s Exponential Smoothing (HWES)

The not-finished complex network implementation can be found in the complex_network branch.

Spatial-only algorithms (original thesis):

• Baseline
  • me.mbe.algorithms.Alg000: Store data on all nodes at all times.
  • me.mbe.algorithms.Alg001: Store data only on closest node when application active.
• Next Node Prediction
  • me.mbe.algorithms.nextnodepred.Alg003: (Multi Order) Markov Model.
  • me.mbe.algorithms.nextnodepred.Alg004: Variable Order Markov Model.
  • me.mbe.algorithms.nextnodepred.Alg008: Store also on some neighboring nodes: Not mentioned in the original thesis
  • me.mbe.algorithms.nextnodepred.Alg012: Fusion Multi Order Markov Model.
• Startup Prediction
  • me.mbe.algorithms.startuppred.Alg011: Store for short pauses.
  • me.mbe.algorithms.startuppred.Alg013: Do not store anything after shutdown: Used together with the algorithms for next node prediction: Not mentioned explicitly in the original thesis
  • me.mbe.algorithms.startuppred.Alg014: Store if short pause predicted.
  • me.mbe.algorithms.startuppred.Alg015: Clustering of startup times for long pauses: Not mentioned explicitly in the original thesis, just as a side note.