Thesis or Dissertation Distributed cloud bursting architecture based on peer-to-peer overlay for service provisioning

Zhygmanovskyi, Andrii

Description
指導教員 : 吉田紀彦
Abstract 11 Introduction 31.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2 Structure of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . 42 Cloud computing and emergence of the Intercloud 63 Peer-to-peer overlay for service sharing and discovery 113.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.2 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133.3 Differences between content and services in peer-to-peer networks . 143.4 Service description . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.5 Service description storing . . . . . . . . . . . . . . . . . . . . . . . 183.6 Service discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193.7 Comparison analysis of (a, v)-graph structure . . . . . . . . . . . . . 224 Peer-to-peer overlay inside the cloud 244.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244.2 Load balancing of service registrations and service queries . . . . . 254.3 Queuing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274.4 Other issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294.5 Simulator implementation . . . . . . . . . . . . . . . . . . . . . . . 294.6 Experiment setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314.7 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334.8 Example domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365 Decentralized architecture for cloud bursting 395.1 Current state-of-the-art . . . . . . . . . . . . . . . . . . . . . . . . . 395.2 Model overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405.3 Service descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 425.4 Service queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425.5 Overlay formation and scaling . . . . . . . . . . . . . . . . . . . . . 435.6 Platform components . . . . . . . . . . . . . . . . . . . . . . . . . . 445.7 Synchronization and conflict resolving in job request/service query queues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475.8 Considerations for evaluation and comparative analysis . . . . . . . 506 Conclusion and future work 536.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536.2 Achievements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536.3 Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546.4 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54Acknowledgments 56Publications 57Bibliography 62
Cloud computing is a technology that has gained extremely wide use in last several years. Initially embraced by major IT companies such as Amazon, Apple, Microsoft, Oracle and Google, which established themselves as top players in the cloud services market, it has became common for most companies to move their infrastructure to the cloud, both public and private. If properly applied, cloud computing not only can help lower IT costs for the enterprise, but also introduces many other benefits, such as effective management of peak-load scenarios by scaling the number of instances according to the real (or predicted) demand, dealing with natural disasters and system outages by seamlessly migrating to other available cloud resources, or serving as a inexpensive platform for the startups with innovative ideas for new services.One of the concerns for the cloud-based solutions is the fact that the components responsible for service discovery, monitoring and load-balancing still employ centralized approaches. The presence of central authority entities like service brokers is often inappropriate, since such solutions lack satisfactory scalability, present a single point of failure and lead to performance bottlenecks and network congestion. On the other hand, considering distributed nature of cloud-based architecture, it is reasonable to use distributed approach to cloud service management and discovery as well, which in turn leads to the idea of using inherently decentralized, fault tolerant and scalable peer-to-peer paradigm.One of ideas that can alleviate already existing and potential problems of centralized cloud is hybrid cloud architecture, that consists in combining both public and private clouds to get more scalable and robust cloud solution. In this thesis, we present an approach to building a hybrid cloud system by employing cloud bursting architecture ― an approach that lies in using external cloud resources when local ones are insufficient. One of major use cases for cloud bursting is managing highly skewed request distribution for deployed services, mostly characterized by peaks in load which are sudden and unpredictable, planned but not long, and often exhibit seasonal behavior.Our cloud bursting architecture is based on the peer-to-peer infrastructure for managing services, which is also an original idea presented in this thesis. This infrastructure is specifically designed to be an efective and scalable solution for storing, sharing and discovering services, and unlike most other peer-to-peer based approaches it allows flexible search queries since all of them are executed against internal database present at each overlay node. We also present several optimizations for peer-to-peer overlay which are necessary for utilizing it in the cloud environment. Evaluation of the peer-to-peer overlay is done by performing a set of experiments on a simulator that show it can be a viable solution to use in cloud setting and specifically in hybrid cloud. We also present some considerations about further cloud architecture evaluation and analysis.Proposed approach is designed to address various issues of interconnecting several clouds, problems of resource provisioning, service deployment and provisioning in the hybrid cloud. Scalability of the approach is attained due to flexibility of service discovery mechanism, decentralized architecture and modular approach, which allows to leverage existing components. We argue that our approach present viable solution for managing abrupt peaks in the load and keeping service provider's QoS and SLA requirements on the desired level.
Full-Text

http://sucra.saitama-u.ac.jp/modules/xoonips/download.php?id=GD0000720

Number of accesses :  

Other information