Programming Language Abstractions for the Global Network

Sibson, Keith (2001) Programming Language Abstractions for the Global Network. PhD thesis, University of Glasgow.

Abstract

Increasing demand for Internet-based applications motivates the development of programming models that ease their implementation. With the research presented in this thesis, we aim to improve understanding of what is involved when programming applications for the global network, and in particular the Web. We are primarily concerned with the development of language-level programming abstractions that address issues arising from the failure and performance properties of the Web. Frequent failure and unpredictable performance are ever-present aspects of any Web computation, so we must bring the properties of the Web into the semantic domain of our program systems. Our primary goal is to enable concise and intuitive expression of failure semantics in the context of concurrency, which is necessary for efficient Web computation given the large overhead in every network access. The main scientific contribution of this thesis is the development of a Web programming model for which a major design goal is the integration of domain concepts, failure interpretation, concurrency, and a mechanism for flow of control after failure. Our model is the first to successfully achieve a clean integration. We develop a programming language called Focus, which incorporates two complimentary abstractions. Persistent relative observables allow reasoning about the dynamic behaviour of computations in the context of past behaviours. Examples of observables are the rate, elapsed time, and success probability of http fetches. The mechanics of our observables mechanism allows the generalisation of the observables concept to all computation, and not just Web fetches. This generalisation is key in our design approach to supervisors, which are abstractions over concurrency designed for the specification of failure semantics and concurrency for computations that contain Web fetches. In essence, supervisors monitor and control the behaviour of arbitrary concurrent computations, which are passed as parameters, while retaining a strict separation of computational logic and control logic. In conjunction with observables, supervisors allow the writing of general control functions, parameterisable both by value and computation. Observables are abstract values that fluctuate dynamically, and all computations export the same set of observables. Observables allow genericity in supervisor control, since the mechanism constrains the value of observables within a pattern of fluctuation around a single number. Whatever the activity of a computation, information about its behaviour can be obtained within a range of values in the observables. This means that supervisors can be applied independently of knowledge of the program logic for supervised computations. Supervisors and observables are useful in the context of the Web due to the multiplicity of possible failure modes, many of which require interpretation, and the need for complex flow of control in the presence of concurrency.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Adviser: Richard Conor
• Keywords: Computer science
• Date of Award: 2001
• Depositing User: Enlighten Team
• Unique ID: glathesis:2001-76088
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 19 Dec 2019 09:15
• Last Modified: 19 Dec 2019 09:15
• URI: https://theses.gla.ac.uk/id/eprint/76088