ActivFORMS: active formal models for self-adaptation

Abstract

Self-adaptation enables a software system to deal autonomously with uncertainties, such as dynamic operating conditions that are difficult to predict or changing goals. A common approach to realize self-adaptation is with a MAPE-K feedback loop that consists of four adaptation components: Monitor, Analyze, Plan, and Execute. These components share Knowledge models of the managed system, its goals and environment. To provide guarantees of the adaptation goals, state of the art approaches propose using formal models of the knowledge. However, less attention is given to the formalization of the adaptation components themselves, which is important to provide guarantees of correctness of the adaptation behavior (e.g., does the execute component execute the plan correctly?). We propose Active FORmal Models for Self-adaptation (ActivFORMS) that uses an integrated formal model of the adaptation components and knowledge models. The formal model is directly executed by a virtual machine to realize adaptation, hence active model. The contributions of ActivFORMS are: (1) the approach assures that the adaptation goals that are verified offline are guaranteed at runtime, and (2) it supports dynamic adaptation of the active model to support changing goals. We show how we have applied ActivFORMS for a small-scale robotic system.

Tele Assistance: A Self-Adaptive Service-Based System Exemplar

“…If desired, additional probes and effectors may be implemented. For the case study, we tested two adaptation solutions realised with ActivFORMS [11], an adaptation engine that executes formally specified MAPE models. The two solutions implemented the following simple strategies: Retry: If a service fails, retry two times Select Reliable: If a service fails, select the equivalent service with the lowest failure rate (and lowest cost if a tie) We used ReSeP probes and effectors including the WorkflowProbe (to track failed service invocations) and the WorkflowEffector (to adapt the workflow as needed).…”

SimCA vs ActivFORMS: comparing control- and architecture-based adaptation on the TAS exemplar

“…ActivFORMS is an architecture-based approach for selfadaptation that uses an integrated formal model of the adaptation components of the feedback loop and the knowledge they share [11]. ActivFORMS distinguishes itself from existing architecture-based approaches in three ways:…”

“…Goal models can be updated at runtime. Here, we do not focus on the goal layer, we refer the interested reader to [11].…”

“…Second, as o✏ine verification is limited to the input provided by the verification models, ActivFORMS allows to continue verification of properties at runtime and inform the user about violations. We refer the interested reader to [11] for details. In the studied TAS scenarios we used both types of runtime guarantees provided by ActivFORMS.…”

“…For this reason adaptation mechanisms based on control theory [10,9] attracted the attention of self-adaptive systems community 1 . Recently, we developed two approaches for runtime adaptation, one from each of the mentioned fields: ActivFORMS (Active Formal Models for Self-adaptation) [11] an architecturebased approach, and SimCA (Simplex Control Adaptation) [16] that is based on principles from control theory and linear optimization. In this paper, we present a comparative evaluation of SimCA and ActivFORMS on a set of scenarios.…”