Section: New Results

Results on Heterogeneous and dynamic software architectures

This year, we focused on the challenges that use models@runtime for resource-constrained and resource-aware systems. Our main results are in the following four subdomains:

• We designed an adaptive monitoring framework for component-based systems in which we highlight the benefits of using models@runtime for adaptive monitoring.

• We improved models@runtime technologies for resource-constrained devices.

• We designed efficient reasoning techniques for dynamic software architecture, focusing in particular on resource consumption optimization challenges.

• We performed several experiments on the Internet of Things application domain.

The next section details our experiments.

Resource-aware dynamic architecture

Modern component frameworks support continuous deployment and simultaneous execution of multiple software components on top of the same virtual machine. However, isolation between the various components is limited. A faulty version of any one of the software components can compromise the whole system by consuming all available resources. We propose a solution to efficiently identify faulty software components running simultaneously in a single virtual machine. It is based on an optimistic adaptive monitoring system to identify the faulty component. Suspected components are instrumented to obtain fined grain data for deeper analysis by the monitoring system, but only when required. Unsuspected components are left untouched and execute normally. Thus, we perform localized, just-in-time monitoring that decreases the accumulated overhead of the monitoring system. We evaluated our approach against a state-of-the-art monitoring system and we have shown that our technique correctly detects faulty components, while reducing overhead by an average of 80% [52] . Based on this work, we have presented two tutorials at the CBSE/QoSA conference [49] and at the Middleware conference [51] .

Technology enablers for resource-aware dynamic software architecture

Models@runtime provides semantically rich reflection layers enabling intelligent systems to reason about themselves and their surrounding context. Most reasoning processes require not only to explore the current state, but also the past history to take sustainable decisions e.g. to avoid oscillating between states. Models@runtime and model-driven engineering in general lack native mechanisms to efficiently support the notion of history, and current approaches usually generate redundant data when versioning models, which reasoners need to navigate. Because of this limitation, models fail in providing suitable and sustainable abstractions to deal with domains relying on history-aware reasoning. This work tackles this challenge by considering history as a native concept for modeling foundations. Integrated in conjunction with lazy load/storage techniques into the Kevoree Modeling Framework, we demonstrated onto a energy-aware smart grid case study that this mechanisms enable a sustainable reasoning about massive historized models [53] .

In this field we also created a specific extension to the docker.io open-source project to support a dynamic resource reservation of running containers [9]

Efficient reasoning techniques for dynamic software architecture

Providing software with the capacity of adapting itself according to its environment requires effective techniques to reason and decide on what adaptation to undertake over the running system. To decide on a system adaptation, we have to characterize the value of the system in its corresponding execution environment. A system cannot be characterized by a single dimension, but only using several dimensions such as performance, energy consumption, security and so on. In this context, we have proposed various techniques to leverage multi-objective evolutionary algorithms both at deployment time [46] , [21] and at runtime  [47] to enable system optimization using multidimensional optimization. We have also proposed a technique to adapt a system proactively based on predictions in order to prevent failures [60]

The Internet of Things application domain

We apply our techniques for heterogeneous and dynamic software architecture more specifically to the Internet of Things application domain. We have two main contributions: (1) an application of the models@runtime concepts on embedded nodes with very limited resources for memory, CPU and battery  [30] , and (2) a study on the problem of renewable energy production and consumption at home [39] . Domestic microgeneration is the onsite generation of low and zero-carbon heat and electricity by private households to meet their own needs. In this paper we explore how an everyday household routine (doing laundry) can be augmented by digital technologies to help households with photovoltaic solar energy generation to make better use of self-generated energy. We present an 8 month in the field study that involved 18 UK households in longitudinal energy data collection, prototype deployment and participatory data analysis [38] . Through a series of technology interventions mixing energy feedback, proactive suggestions and direct control, the study uncovered opportunities, potential rewards and barriers for families to shift energy consuming household activities. The study highlights how digital technology can act as a mediator between household laundry routines and energy demand-shifting behaviors. Finally, the study provides insights into how a “smart” energy-aware washing machine shapes organization of domestic life and how people “communicate” with their washing machine.