Dynamic grid load sharing with adaptive dissemination protocols
The Journal of Supercomputing
Autonomic cloud resource sharing for intercloud federations
Future Generation Computer Systems
| Hi-index | 0.01 |
As grids scale to include more individual resources, improved grid middleware services that eliminate the need for structure and centralization becomes more important. Moreover, the most effective grid middleware services will be adaptive, reacting to their highly dynamic environments. One important service that must be distributed and scalable is grid resource scheduling. In hybrid grids, resources and jobs are likely to be non-uniformly distributed in space and in time, and thus no single approach to tracking resource information will be effective in all places at all times. Non-uniform information dissemination helps enable large-scale and dynamic grid resource scheduling, with less packet overhead and better localized coverage, which is especially useful when the grid structure is mostly unknown, such as in hybrid grids. This dissertation describes improved information dissemination protocols that adapt to their environment using feedback from the system. Thus, grid nodes produce different dissemination protocols on-the-fly, where each protocol individually reflects both the characteristics of particular resource and load distributions, and the policies of autonomous grid nodes or regions. The adaptive information dissemination protocols can result in much less packet overhead, with comparable query satisfaction rates, compared to best-case non-adaptive protocols that may be configured for each particular grid resource and load scenario.