Run-time Management Framework Research Articles

BarMan: A run-time management framework in the resource continuum

Over the last years, the number of IoT devices has grown exponentially, highlighting the current Cloud infrastructure limitations. In this regard, Fog and Edge computing began to move part of the computation closer to data sources by exploiting interconnected devices as part of a single heterogeneous and distributed system in a computing continuum viewpoint. Since these devices are typically heterogeneous in terms of performance, features, and capabilities, this perspective should encompass programming models and run-time management layers. This work presents and evaluates the BarMan open-source framework by implementing a Fog video surveillance use-case. BarMan leverages a task-based programming model combined with a run-time resource manager and the novel BeeR framework to deploy the application’s tasks transparently. This enables the possibility of considering aspects related to the energy and power dissipated by the devices and the single application. Moreover, we developed a task allocation policy to maximize application performance, considering run-time aspects, such as load and connectivity, of the time-varying available devices. Through an experimental evaluation performed on a real cluster equipped with heterogeneous embedded boards, we evaluated different execution scenarios to show the framework’s functionality and the benefit of a distributed approach, leading up to an improvement of 66% on the frame processing latency w.r.t. a monolithic solution.

Reliability Aware Design and Lifetime Management of Computing Platforms

Meeting reliability targets with viable costs in the nanometer landscape become a significant challenge, requiring to be addressed in an unitary manner from design to run time. To this end, we propose a holistic reliability-aware design and lifetime management framework concerned (i) at design time, with providing a reliability enhanced adaptive architecture fabric, and (ii) at run time, with observing and dynamically managing fabric's wear-out profile such that user defined Quality-of-Service requirements are fulfilled, and with maintaining a full-life reliability log to be utilized as auxiliary information during the next IC generation design. After introducing our framework and the general philosophy behind it we delve into its key components. Specifically, we first introduce design time transistor and circuit level aging models, which provide the foundation for a 4-dimensional Design Space Exploration (DSE) meant to identify a reliability optimized circuit realization compliant with area, power, and delay constraints. Subsequently, to enable the creation of a low cost but yet accurate fabric observation infrastructure, we propose a methodology to minimize the number of aging sensors to be deployed in a circuit and identify their location, and introduce a sensor design able to directly capture circuit level amalgamated effects of concomitant degradation mechanisms. Furthermore, to make the information collected from sensors meaningful to the run-time management framework we introduce a circuit level model that can estimate the overall circuit aging and predict its End-of-Life based on imprecise sensors measurements, while taking into account the degradation nonlinearities. Finally, to provide more DSE reliability enhancement options we focus on the realization of reliable processing with unreliable components, and propose a methodology to obtain Error Correction Codes protected data processing units with an output error rate smaller than the fabrication technology gate error rate.

ARTE: An Application-specific Run-Time managEment framework for multi-cores based on queuing models

Abstract This paper presents an Application-specific Run-Time managEment (ARTE) framework to tackle the problem of managing computational resources in an application specific multi-core system. The ARTE framework run-time goal is to minimize applications’ response times while meeting the applications’ computational demands and fitting within the available power budget. The approach addresses application specific embedded systems assuming that the set of target applications is known at design-time. In addition, it considers run-time scenarios that are unpredictable due to variable user activity and/or interaction with the external environment. ARTE takes decisions about resource distribution to the active applications at run-time once the system state is known. ARTE leverages an analytical queuing model at run-time to predict the applications’ response times. The accuracy of this model is enhanced by accounting for contention overhead on the resources shared among the active applications. The analytical nature of the queuing model allows an estimation of the system performance with negligible overhead. Finally we compared the proposed ARTE framework to state-of-the-art techniques to assess its benefits in terms of systems performance and run-time overhead for the selected set of parallel benchmarks.

Synchronizing concurrent model updates based on bidirectional transformation

Model-driven software development often involves several related models. When models are updated, the updates need to be propagated across all models to make them consistent. A bidirectional model transformation keeps two models consistent by updating one model in accordance with the other. However, it does not work when the two models are modified at the same time. In this paper we first examine the requirements for synchronizing concurrent updates. We view a synchronizer for concurrent updates as a function taking the two original models and the two updated models as input, and producing two new models where the updates are synchronized. We argue that the synchronizer should satisfy three properties that we define to ensure a reasonable synchronization behavior. We then propose a new algorithm to wrap any bidirectional transformation into a synchronizer with the help of model difference approaches. We show that synchronizers produced by our algorithm are ensured to satisfy the three properties if the bidirectional transformation satisfies the correctness property and the hippocraticness property. We also show that the history ignorance property contributes to the symmetry of our algorithm. An implementation of our algorithm shows that it worked well in a practical runtime management framework.

Investigating Autonomic Runtime Management Strategies for SAMR Applications

Dynamic structured adaptive mesh refinement (SAMR) techniques along with the emergence of the computational Grid offer the potential for realistic scientific and engineering simulations of complex physical phenomena. However, the inherent dynamic nature of SAMR applications coupled with the heterogeneity and dynamism of the underlying Grid environment present significant research challenges. This paper presents application/system sensitive reactive and proactive partitioning strategies that form a part of the GridARM autonomic runtime management framework. An evaluation using different SAMR kernels and system workloads is presented to demonstrate the improvement in overall application performance.