Wireless networks and inexpensive sensors are leading to a world rich in geographically distributed sources of data. Various massively networked systems are being designed to manage this information. Such systems typically contain two key architectural components: a software ""network"" connecting the participating nodes in the system, and a query engine running across the nodes. There is little consensus on how to co-engineer these components to achieve an efficient, robust solution. However, recent work on Declarative Networking showed that both these layers can be naturally implemented in a high-level language, in which the programmer specifies ""what"" result is desired, not ""how"" it is to be achieved.
This project develops ""Dynamic Metacompilation"" to enable automatic co-optimization of the network and query components of such systems. In this framework, the optimizing compiler for a networked information system is bootstrapped from optimization rules written in the very same declarative, distributed language that is being optimized. Such a scheme will bring remarkable ease of software engineering for this new class of systems.
The results of the work can significantly accelerate the development of a new generation of systems to harness the potential of a massively networked and data-rich world, with implications for a broad range of applications including environmental monitoring, public health, disaster response, and computer system management. Project funds support the training and research of two graduate students, and plans are underway for use of the software in coursework. Publications, technical reports, software and experimental data from this research will be disseminated via the project web site (http:// p2.cs.berkeley.edu).
