Savasana: Adaptation Consistency via Code Analysis

Savasana is the first white-box approach that uses code analysis for reasoning about consistency of adaptation.

Approach Overview

Savasana consists of two parts: Static Code Analysis runs on the system's code and Run-time Control manages the corresponding running system.

We realized Savasana's code analysis capabilities on top of Soot. Soot is a Java optimization framework, which can also be used for static analysis. We used Soot for the implementation of Dependency Extractor and Model Analyzer. We implemented Savasana's runtime control capabilities on top of the Spring framework. For implementing the Monitor component, we used Spring's extensive support for aspect-oriented programming, and for the Adapter component, we relied on Spring's dynamic (re)con.figuration facilities.

Implementation

Savasana project source code is available here.

Evaluation

We examine Savasana's behavior by measuring two properties:

• Reachability: the amount of time it takes for fulfilling the adaptation request.
• Disruption: the amount of time that application threads are suspended for adaptation.

There are two factors that impact these properties:

• The internal structure of root transactions (i.e., scenarios).
• The level of concurrency in the system.

The experiments are conducted on a software system called Emergency Deployment System (EDS), which is intended for the deployment and management of personnel in emergency response scenarios.
EDS source code is available here.

We applied Savasana in three scenarios, each one is adopted from one of EDS's main tasks, namely: Strategy Analysis, Deployment Advising, and Resource Estimation. Inter-component dependency of these scenarios are depicted in the following figure.
The evaluation results are available here.

Publications

More details about Savasana can be found in our publication:

• Ensuring the Consistency of Adaptation through Inter- and Intra-Component Dependency Analysis
  Alireza Sadeghi, Naeem Esfahani, and Sam Malek
  ACM Transactions on Software Engineering and Methodology (TOSEM)