Study, Performance Evaluation and Customization of Grid Service Oriented Architecture for Satellite Image Processing by Nitant Dube

Guided by: Dr. K. S. Dasgupta With Synopsis and CD 05EXTPHDE04

ABSTRACT:

Recent developments in space technology and exponential increase in demand of earth observation data has generated a need to study software architecture, carry out performance evaluation, propose and validate customized software framework for processing, analysis and dissemination of earth observation data.

Grid Services for Earth Observation Image Data Processing (GEO-ID) proposed as part of this research work is a software framework for user centric data processing, with a goal of extensively using grid and service oriented architecture.

Grid Process Scheduling Service (PSS) developed as part of this research work helps in meeting performance and functional requirements of satellite image processing. Proposed customizations help in improving performance by scheduling products on remote grid nodes based on scenarios such as location of raw satellite data, CPU utilization and I/O loads of the participating nodes. Processing performance of a node is improved by modelling process sequence as Direct Acyclic Graph (DAG) and adaptive queue based scheduling.

Data dependent techniques of process scheduling provide capability of scheduling processes on nodes, where data resides. These techniques fail for cases, where a single workflow requires access to computational resources not available on a single system. Applications such as archive mode of processing, where all received data at the ground station is required to be processed as part of standard data processing chain falls under this category. The augmented GEO-ID architecture provides capability of using the underlying distributed operating system for execution of grid services.

An adaptive data transfer model is also proposed as part of this research work. The proposed adaptive model is an extension of GridFTP, which is a default protocol used for transfer of data on grid. Adaptive data transfer model utilizes image characteristics in conjunction with the architectural feature including network capability for performance enhancement.

Validation of the proposed GEO-ID software framework is carried out by executing three earth observation applications on the grid test bed. These three applications include time series processing, near real time tracking of events and multi sensor data fusion. These applications cover broad categories of earth observation data users. Results obtained by executing these three applications on GEO-ID are compared with the results of execution of these applications using conventional architecture.

Satellite data volumes in future are going to exponentially increase due to higher spatial and spectral resolutions. More user specific processing requirements are going to come in future and hence we see that GEO-ID and the research carried out on the proposed area for performance improvement has good scope in future.