Resampling routine on heterogeneous networked-based environment

In this paper, we have described a model to parallelize the resampling routine, which is used in the geometric correction of data provided by remote sensing satellites. Our model is a typical master-slave model consisting of N machines termed as hosts out of which one is designated as the master. The input image data resides on the master. Processing of the input image data is done in parallel on the N machines. Issues related to load-balancing and various error conditions that may occur during execution like one of the machines going down have been studied and are incorporated in the model. It also provides the flexibility to add or delete the hosts during the execution of the resampling routine. The serial version of this routine involves huge amount of computations and takes substantial amount of time even for an image of 473 MB. We have implemented our model with the help of PVM which is most often used in distributed computing environment. Our approach has been tested for geometric correction on LISS-III 4 band data of size 473 MB. It is found that if one uses 2, 3 or 4 hosts the overall execution time is reduced by 33%, 42% and 49%, respectively.