ST-CRTS: Enabling Processing of Large-Scale Scientific Data Through Compiler Supported XML Abstractions

Ohio State University Research Foundation -Do Not Use, Columbus OH

Investigators

Abstract

Background Analysis of large and/or geographically distributed scientific datasets, is emerging as a key component of grid computing and holds great promise for scientific discoveries. However, such analysis tasks are complicated by low-level and specialized data storage formats, and lack of standard interfaces for accessing the data. Scientific datasets are typically stored in binary or character flat-files. These enable compact storage and efficient processing, but, make the specification of processing much harder. The overall motivation for this project is "future distributed and grid computing applications will need high-level abstractions and programming languages to access, retrieve, and process large-scale scientific datasets that are physically stored in compact low-level formats". Intellectual Merit Develop a compiler to support a system of the nature described above involves a number of challenges. These challenges arise primarily because of the following four reasons: 1) Supporting high-level abstractions, which implies that compiler not only needs to generate code that will correctly access data, but also needs to automatically achieve high-locality 2) Large Disk-Resident Datasets, which makes it quite challenging to generate low-level code that will have high data access locality 3) Characteristics of the Applications we are targeting scientific data analysis, data mining, and image analysis applications that have not received much attention in the restructuring compilers community, 4)Features of XQuery since XQuery is derived from functional, declarative, as well as database programming languages, a number of issues arise because of the constructs and programming styles it supports. Broader Impact -The PI is working on creating a new two quarter sequence on application-driven grid computing. Here, one of the topics that will be emphasized is the analysis of large datasets arising from real biological and medical applications. - The PI is currently working with three female Ph.D students, and expects to involve one of them in this project. Together with many of his colleagues in the Computer Science and Engineering department at the Ohio State University, Agrawal is part of a collaboration plan with two HBCUs in the state of Ohio. This collaboration plan includes: 1) Guest lectures by OSU faculty and 2) Research opportunities for HBCU undergraduates through independent study and summer projects jointly with OSU faculty (we hope to be able to use NSF REU supplements for funding this activity).