D6.2. Final Workload, Load Propagation, and Application Models

RECAP's work package WP6 targets application modelling, workload modelling, and workload propagation modelling, and this deliverable aims to provision an extension of the models presented in (D6.1). The achievements of the WP are demonstrated throughout the document with the following outline:

1. The approaches and theoretical/analytical results of the application characterisation;
2. A detailed description of workload models for use cases, a comparison of these models, and their utilisation of models;
3. A detailed description of workload propagation models devised based on application models presented in Deliverable D6.1, and an implementation of these models and application of the models to use cases; and
4. The approaches and preliminary results of practical application of the models.

The main parts of the document demonstrate workload models and workload propagation models for use cases, which are developed and constructed using different techniques ranging from the basic probability-based, analytical, to machine-learning techniques. While workload models illustrate what the load looks like over time and are also used for future load predictions, workload propagation models capture the traffic flows in the systems. Both are integrated for synthetic workload generations. In addition, workload predictions and workload propagation also provide a base for reasoning on elasticity and remediation of systems.

With the complementary and extended models, the work packages provides a baseline for other WPs. More specifically, workload models are used to realise workload predictors. Predictors and application optimisers in WP6 are a core building block of the entire RECAP architecture (D4.6); workload propagation models are the bases of the algorithms for synthetic workload generation (D5.3); and the approaches and models of elasticity bring up some instructions/guidelines for both the development and improvement of schemes of autoscaling and optimisation.