Research On Grid Data Transferring Technologies Supporting Dynamic Updating

Data-intensive scientific and engineering applications require the transfer of large amount of data in grid, in which process the performance and efficiency are two important indicators. This thesis researches on grid data transferring and its dynamic updating technologies. On one hand, two data transferring approaches are proposed, both of which can enhance the transferring performance. On the other hand, dynamic updating of grid data transfer is researched and solution is proposed. The main contents of the thesis include the following:(1) This thesis proposes striped transferring approach based on multiple replicas, which performs data transferring by simultaneously using multiple replicas and shielding the protocol differences and domain differences among replicas. Experiments prove that the data transferring performance is enhanced while adopting the approach.(2) When grid applications operate on multidimensional data, query and transfer are two widely used operations. To support flexible query operation, a storage structure based on partition the multidimensional data is adopted. However, this will cause decrease of the transfer performance. This paper investigates this problem and proposes the logically linked file (LLF) approach to solve it. By virtually linking the qualified non-continuous data sets to a LLF and providing a specific transfer mechanism for LLF, the transfer performance is improved.(3) Mission-critical applications in grid require the online evolvement ability of the data transferring software, namely, the ability to dynamic update the software while the transfer is ongoing. This thesis proposes a solution to the problem, including a dynamic updatable software composing specification named the Du-OSGi Specification, a dynamic update framework named the Du-OSGi Framework, and a set of dynamic updating tools. After applying the solution to data transferring, the data transferring software acquires the online evolvement ability. Experiments show the data transferring performance will not be influenced by updating.