Autonomic cloud resource sharing for intercloud federations

摘要

Recent advances in information technology make remote collaboration and resource sharing easier for next generation distributed systems, such as grids and clouds. One common model of study is the convergence of these systems, along with interclouds to a unified global computing resource. Despite similarities between grids and clouds, there are a number of fundamental differences that make this convergence process harder. For example, clouds have inherent administrative boundaries, something which the grid computing paradigm avoided from the early stages of research. Such administrative boundaries primarily affect capabilities of clouds to be interoperable. Moreover, they also negatively affect the possibility of a seamless intercloud federation on the path to convergence. Resource sharing in general and related communication methodologies, such as information dissemination and matchmaking are also integral elements in this convergence process. To help improve the success of distributed cloud resource schedulers, we propose proxies that disseminate information as agents of dissemination sources. Such proxies can then make information about resource states available at 'distant' clouds, where there may be no direct, or even no indirect control. Moreover, they can make this resource state available more efficiently than where no proxies are used. In particular, with proxies, dissemination overhead is reduced by up to 65% under different scenarios, where existing solutions may not even produce efficient protocols. In addition, proxies help reduce dissemination overhead by 19% on average. Our results also show that randomly selecting proxy nodes perform comparably to other strategies that may select proxies based on particular criteria.