• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권9호
• /
• pp.4280-4300
• /
• 2017

The fact that cloud computing services have been proposed in recent years, organizations and individuals face with various challenges and problems such as how to migrate applications and software platforms into cloud or how to ensure security of migrated applications. This study reviews the current challenges and open issues in cloud computing, with the focus on autonomic resource management especially in federated clouds. In addition, this study provides recommendations and research roadmaps for scientific activities, as well as potential improvements in federated cloud computing. This survey study covers results achieved through 190 literatures including books, journal and conference papers, industrial reports, forums, and project reports. A solution is proposed for autonomic resource management in the federated clouds, using machine learning and statistical analysis in order to provide better and efficient resource management.

• 한국정보전자통신기술학회논문지
• /
• 제17권4호
• /
• pp.213-220
• /
• 2024

본 논문에서는 WebAssembly(WASM)를 활용하여 디바이스와 엣지 클라우드 간의 Federated Learning을 수행하는 최적화 전략을 제안한다. 제안된 전략은 일부 학습을 디바이스에서 수행하고, 나머지 학습을 엣지 클라우드에서 수행하여 효율성을 극대화하는 것을 목표로 한다. 특히, GPU 메모리 세그먼트 간 데이터 이동과 연산 작업의 중첩을 최적화하여 전체 학습 시간을 줄이고 GPU 사용률을 향상시키는 방법을 수학적으로 설명하고 평가한다. 다양한 실험 시나리오를 통해 비동기 데이터 전송과 연산 중첩이 학습 시간을 단축하고 GPU 사용률을 향상시키며, 모델 정확도를 증가시킴을 확인하였다. 모든 최적화 기법을 적용한 시나리오에서 학습 시간이 47% 단축되었고, GPU 사용률은 91.2%로 향상 되었으며, 모델 정확도는 89.5%로 증가함을 확인하여 비동기 데이터 전송과 연산 중첩이 데이터 전송을 기다리는 GPU 유휴 시간을 줄이고, 병목 현상을 완화할 수 있음을 확인하였다. 본 연구는 향후 Federated Learning 시스템의 성능 최적화에 기여할 수 있을 것으로 사료된다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제10권11호
• /
• pp.5271-5289
• /
• 2016

On large-scale data analysis platforms deployed on cloud infrastructures over the Internet, the instability of the data transfer time and the dynamics of the processing rate require a more sophisticated data distribution scheme which maximizes parallel efficiency by achieving the balanced load among participated computing elements and by eliminating the idle time of each computing element. In particular, under the constraints that have the real-time and limited data buffer (in-memory storage) are given, it needs more controllable mechanism to prevent both the overflow and the underflow of the finite buffer. In this paper, we propose an auto regulated data provisioning model based on receiver-driven data pull model. On this model, we provide a synchronized data replenishment mechanism that implicitly avoids the data buffer overflow as well as explicitly regulates the data buffer underflow by adequately adjusting the buffer resilience. To estimate the optimal size of buffer resilience, we exploits an adaptive buffer resilience control scheme that minimizes both data buffer space and idle time of the processing elements based on directly measured sample path analysis. The simulation results show that the proposed scheme provides allowable approximation compared to the numerical results. Also, it is suitably efficient to apply for such a dynamic environment that cannot postulate the stochastic characteristic for the data transfer time, the data processing rate, or even an environment where the fluctuation of the both is presented.

• Journal of Communications and Networks
• /
• 제13권6호
• /
• pp.567-582
• /
• 2011

The growing popularity of the Internet has caused the size and complexity of communications and computing systems to greatly increase in recent years. To alleviate this increased management complexity, novel autonomic management architectures have emerged, in which many automated components manage the network's resources in a distributed fashion. However, in order to achieve effective collaboration between these management components, they need to be able to efficiently exchange information in a timely fashion. In this article, we propose a context dissemination framework that addresses this problem. To achieve scalability, the management components are structured in a hierarchy. The framework facilitates the aggregation and translation of information as it is propagated through the hierarchy. Additionally, by way of semantics, context is filtered based on meaning and is disseminated intelligently according to dynamically changing context requirements. This significantly reduces the exchange of superfluous context and thus further increases scalability. The large size of modern federated cloud computing infrastructures, makes the presented context dissemination framework ideally suited to improve their management efficiency and scalability. The specific context requirements for the management of a cloud data center are identified, and our context dissemination approach is applied to it. Additionally, an extensive evaluation of the framework in a large-scale cloud data center scenario was performed in order to characterize the benefits of our approach, in terms of scalability and reasoning time.