Structured Orchestration of Data and Computation
Presented as part of the 2012 HCSS conference.
Abstract:
This talk will describe our attempt to integrate data and services within a large organization, or to use the Internet. We have developed a theory, called Orc, in which we specify the sources of data and computations and how to orchestrate their executions (concurrently, one after the other, only when the majority of data is available, ...). The Orc programming language has a very small kernel, using external libraries of sites for support functions. The language starts with concurrency as the defining characteristic of programs. With its hierarchical and recursive combination facilities, it provides a model of structured concurrency. It can be used to orchestrate a small application or a giant one with many millions of threads, running in real time, from milliseconds to months. We have developed an experimental implementation of the language. The language, its documentation, and a web-interface are available at www.orc.csres.utexas.edu where programs can also be submitted for execution.
Motivation for the Work
The Internet promises to improve our lives by allowing us to connect to data and services that support our daily existence. It promises to improve business processes by integrating data from a variety of sources within specific applications. This vision has not yet been fully realized. Instead, our infrastructure is currently built from numerous closed-world systems, each built by a single individual or business entity, which enclose and protect data and computation, while providing access to their services through limited human-oriented interfaces. As a result, users are forced to manually mediate between these services, thus wasting time and effort.
To illustrate this situation, consider the following scenario. A patient receives an electronic prescription for a drug from a doctor. The patient compares prices at several near-by pharmacies, and chooses the cheapest one to fill the prescription. He pays the pharmacy and receives an electronic invoice which he sends to the insurance company for reimbursement with instructions to deposit the amount in his bank account. Eventually, he receives a confirmation from his bank. The entire computation is mediated by the patient who acquires data from one source, does some minor computations and sends data to other sources. The mediation is entirely manual.
This computing scenario is repeated millions of times a day in diverse areas such as business computing, e-commerce, health care and logistics. In spite of the extraordinary advances in mobile computing, human participation is currently required in every major step in most applications. This is because the infrastructure for efficient mediation is largely absent, thus contributing to cost and delay in these applications. We believe that humans can largely be eliminated, or assigned a supporting role, in many applications. Doing so is not only beneficial in terms of efficiency, but also essential if we are to realize the full potential of the interconnectivity among machines, using the services and data available in the internet or within a large organization. We would prefer that humans advise and report to the machines, rather than that humans direct the machines in each step. Orc has proved successful in this effort because it has small number of key concepts and a strong theoretical basis.
Biography:
Jayadev Misra is a professor and holder of the Schlumberger Centennial chair in Computer Science at the University of Texas at Austin. He is known for his work in the area of concurrent programming, with emphasis on rigorous methods to improve the programming process. His work on the UNITY methodology, jointly with Chandy, has been influential in both academia and industry, and has spawned a large number of tools and research projects. He has recently developed a programming language, called "Orc", for concurrent orchestrations of interacting components. He is also spear-heading an effort, jointly with Tony Hoare, to automate large-scale program verification. Misra is a fellow of ACM and IEEE. He held the Guggenheim fellowship during 1988-1989. He was the Strachey lecturer at Oxford University in 1996, and has held the Belgian FNRS International Chair of Computer Science in 1990. He is a member of the Academy of Distinguished Teachers at the University of Texas at Austin. Misra has been the past editor of several journals including: Computing Surveys, Journal of the ACM, Information Processing Letters, and the Formal Aspects of Computing. He is the author of two books, Parallel Program Design: A Foundation, Addison-Wesley, 1988, co-authored with Mani Chandy, and A Discipline of Multiprogramming, Springer-Verlag, 2001.