Pages: pp. 1-2
Cloud of Clouds is viewed as the next revolution in the Cloud Computing paradigm wherein the computational and data infrastructure for handling scientific, business and enterprise applications spans across multiple clouds and Data-Centers (DCs). Since it was introduced, Cloud Computing research has addressed a range of issues from realizing the physical infrastructure and its virtualization, to providing a range of services (PaaS, SaaS, IaaS) to Cloud users and building Cloud applications. The Cloud paradigm is also increasingly being used to support more traditional high-performance applications.
Predominantly a Cloud platform offers adequate level of transparency to let users work on their application dependent data and programs rather than pinning them from underlying infrastructure related challenges. Further, there seems to be a sustained interest in porting high-performance applications support on Clouds, which calls for an entire set of new challenges to be addressed. This is primarily because high-performance computing platforms despite allowing resource scaling distinctly enjoys a tightly coupled configurations that minimize all overheads and communication delays whereas, on a Cloud platform providing a high-speed communications support is a challenging task to match the speeds of a HiPC platforms. Nevertheless, at least, for nontime-critical applications overhead and delay management outweighs when compared to monetary and power saving benefits.
As the complexity, heterogeneity and scale of applications grows, it will be increasingly important to be able to compose federated Cloud of Clouds platforms that can address the need for heterogeneous capabilities and large scale capacities. For example, with an anticipated growth of mobile users of Cloud services in the near future, issues related to the interoperability between Cloud service providers, Cloud technologies and users become challenging. Business ventures operating across multiple Clouds may need to set and comply with policy driven schemes, which may become prohibitively expensive for ordinary users. In addition, pricing models, trust and security based research are added issues that need to be addressed.
IEEE Transactions on Computers has recognized this important and timely issues in this special issue. This special issue is motivated by these compelling requirements and challenges, and aims to compile research that aim to fundamentally address them. It has attempted to cover a broad range of research issues that directly influence realizing and effectively using a Cloud-of-Clouds platform. We received an overwhelming response from the community of researchers, including both system theorists as well as practitioners. The accepted papers reflect the diversity of the research areas underlying the Cloud of Clouds paradigm and can be grouped under six different and distinct categories, viz., infrastructure based studies, resource procurement strategies, data storage, energy/power management, cloud as a market place, and finally an application driven study using the MapReduce paradigm. Collectively, the six selected papers attempt to strike a balance between systems theory and practical utility in their respective contributions.
The first paper, “Data Similarity-Aware Computation Infrastructure for the Cloud,” by Yu Hua, Xue Liu, and Dan Feng, is based on infrastructure aspects wherein they clearly elicit the need for looking into alternate infrastructure possibilities employing multicore machines and evaluating the performance. This is also driven by the fact that in the current day implementations multilevel cache hierarchy is becoming more important to obtain satisfactory performance on the cloud. To this end, they propose a novel architecture, an efficient and cost-effective multilevel caching scheme, that is shown to be completely flexible and able to handle fairly large volume data with minimal data migration across nodes, which is an imperative step towards realizing a Cloud of Clouds platform. The authors perform an important step of benchmarking against several known and commonly used schemes w.r.t several performance metrics such as hit ratio, data migration, throughput, time complexities, query processing times, etc.
The second paper, “A Mechanism Design Approach to Resource Procurement in Cloud Computing,” by Abhinandan S. Prasad and Shrisha Rao proposes three different strategies to automate resource procurement and authors demonstrate over multiple service providers, a key aspect in a Cloud of Clouds set up. For the three strategies authors present a rigorous analysis that derives the payment rules and they validate the strategies using a broker model that automates across different cloud vendors. This study will be an important contribution when real-life implementation on a multicloud vendor set up is carried out.
The third paper, “NCCloud: A Network-Coding-Based Storage System in a Cloud-of-Clouds,” by Henry C.H. Chen, Yuchong Hu, Patrick P.C. Lee, and Yang Tang, takes a new step towards proposing a novel architecture, referred to as NCCloud, that is cost effective for repairing and recovering data in a multiple cloud set up. NCCloud is designed based on FMSR codes, which are shown to guarantee the same level of fault tolerance and data redundancy as in conventional erasure codes, however a distinct advantage comes from the fact that FMSR codes seem to incur less monetary cost due to minimal data transfer. Authors demonstrate the performance via cost and reliability analysis. The proposed approach is certainly an attractive solution to handle modern day large scale data that uses multiple cloud storage system using proxies.
The fourth paper, “Optimal Power Allocation and Load Distribution for Multiple Heterogeneous Multicore Server Processors across Clouds and Data Centers,” by Junwei Cao, Keqin Li, and Ivan Stojmenovic, presents an interesting problem of optimizing power and load distribution across multiple data centers, which is an important issue when considering a cloud of clouds set up. Specifically authors consider heterogeneous infrastructure wherein servers have multicore CPUs and formulate a power constrained optimization problem that is shown to utilize the available resources efficiently. Rigorous theoretical results are presented towards demonstrating certain key results that lead to design decisions. Energy efficiency being an important metric to consider the problem addressed is timely for a Cloud of Clouds set up.
While the above papers address certain specific technical aspects on infrastructure and resource utilization aspects, an interesting formulation is presented in the fifth paper, “Price Competition in an Oligopoly Market with Multiple IaaS Cloud Providers,” by Yuan Feng, Baochun Li, and Bo Li, in dealing with pricing aspects, specifically when multiple IaaS Cloud Providers are in place. Authors approach the problem by formulating as a noncooperative game and derive sufficient conditions for the existence of a Nash equilibrium. In certain special case of interest authors extend the analysis to characterize the equilibrium state. Based on the rigorous analytical results authors propose an iterative algorithm to evaluate the performance of their findings. Several observations from Cloud users' perspective and also from Service Providers perceptive are derived which makes this contribution a value-added one when migrating to a cloud-of-clouds infrastructure.
Finally, the sixth paper, “From the Cloud to the Atmosphere: Running MapReduce across Data Centers,” by Chamikara Jayalath, Julian Stephen, and Patrick Eugster, considers an interesting application problem using MapReduce paradigm across multiple data centers. Authors consider geo-distributed datasets which are large-scale data that are distributed across data centers. Till date most literature considers solutions using MapReduce paradigm when datasets dwell in single data centers. This paper attempts to deal with executing sequences of data sets and propose methodology to aid efficient scheduling of such job sequences with minimum execution time and monetary cost. Authors introduce a novel G-MR system to execute such sequence of jobs and present empirical evidence form real-life Cloud service providers.
Although the selected papers contribute and address only a limited set of issues related to Cloud of Clouds, the papers clearly demonstrate the need for such a technology and also highlight underlying issues and challenges. We would like to thank all the researchers who had responded to this call and acknowledge the efforts that went into their submissions. We also thank the Editor-in-Chief Professor Albert Zomaya for supporting this timely topic as a special issue. Finally, we thank Ms. Carrie James, administrator for IEEE TC, for her constant support without which this special issue would not have been completed in time!