SOFTWARE: High-level Programming Methodologies for Data Intensive Computations

点击翻译按钮获取中文摘要

Solving problems that have large computational and storage requirements is becoming increasingly critical for advances in many domains of science and engineering. Runtime and compiler technology can significantly accelerate advances in such domains by enabling the programming of algorithms for these problems at very high levels of abstraction. Our research focuses on an important class of scientific and engineering problems: data intensive computations. Such computations arise in many domains of scientific and engineering research. We are targetting two high-level programming models for this class of computations:1. Object-Oriented (Java-Based): Object-oriented features like encapsulation and polymorphism can ease software development and allow better long-term management of large computational projects. We are using a dialect of Java for expressing data intensive computations. An initial prototype compiler for this dialect of Java has been built. Our compiler uses static interprocedural slicing technique to extracting required functions for utilizing an existing runtime system, Active Data Repository (ADR). 2. Declarative (XQL-Based): The very high-level nature of declarative languages can allow rapid prototyping of new computational techniques. XML (Extensible Markup Language) is getting accepted as the format for the logical layout of data which may be made available over the web or exchanged between organizations. We will make extensions to XML Schemas for expressing multidimensional collections of elements. We also will also make extensions to XQL, which is a query language for XML documents, for expressing processing of large datasets. We propose innovative research in five important areas:1. Execution Strategies for Disk Resident Datasets. 2. Advanced Symbolic Analysis. 3. Interprocedural Code Motion Techniques for Object-Oriented Programs. 4. Combined Compiler and Runtime Techniques for Locality Management. 5. Advanced Interprocedural Slicing Techniques for Object-Oriented Languages.