Resources in the Grid exhibit different availability properties and patterns over time, mainly due to their administrators’ policies for the Grid, and the different domains to which they belong, e.g. non-dedicated desktop Grids, on-demand systems, P2P systems etc. This diversification in availability properties makes availability-aware resource selection, for applications with different fault tolerance capabilities, a challenging problem. To address this problem, we introduce new availability metrics for resource availability comparison. We further predict resource availability considering their availability policies. We introduce a new resource availability predictor based on pattern matching through availability pattern recognition and classification for resource instance and duration availability, and compare it with other methods. Notably we are able to achieve an average accuracy of more than 80% in our predictions.

The technology that significantly supports the load-balancing and fault-tolerance capabilities in Grids is the job’s checkpointing mechanism. Nevertheless, contemporary Grid environments distinctly lack the possibility of integration with the low-level processes’ checkpointers. Nowadays, some Grids support the checkpointing of applications which internally implement this functionality and additionally adhere to the imposed interface. On the contrary, the paper describes the Grid Checkpointing Service, the prototype design and implementation of Grid-level service, which makes it possible to utilize low-level or legacy checkpointing packages in Grids. In fact, the presented service is a proof-of-concept implementation of a part of the Grid Checkpointing Architecture (GCA). Nevertheless, the way the GCS was implemented allows it to be installed and utilized independently of the other parts of the GCA.