Department of Computer Science and Technology

Technical reports

Distributed computing with objects

David Martin Evers

March 1994, 154 pages

This technical report is based on a dissertation submitted September 1993 by the author for the degree of Doctor of Philosophy to the University of Cambridge, Queens’ College.

DOI: 10.48456/tr-332

Abstract

Distributed systems and object-based programming are now beginning to enter the mainstream of computing practice. These developments have the potential to simplify the distributed application programmer’s task considerably, but current systems impose unnecessary burdens. Distributed operating systems provide palatable message passing between remote processes but leave the preparation and interpretation of the messages to application code. Remote procedure call systems use familiar language-level concepts to hide distribution, but the awkwardness of service creation and binding discourages the use of transient objects. Finally, object-based programming languages which support distribution often ignore the possibility of failures and do not efficiently accommodate heterogeneity.

This dissertation discusses the design, implementation and evaluation of a practical system for network objects which addresses these problems for a representative programming language (Modula-3) and distributed computing environment (the ANSA testbench). We propose that language level objects should explicitly represent bindings to potentially remote access points (interfaces), which are sufficiently lightweight that they can be used as transient handles for shared state. Our system uses local objects to stand for remote services and local method call to cause remote operation invocation. Within a process, concurrency control is provided by familiar language-level facilities. The local programming language’s object type system is made to represent the global service type system in a natural way. We support dynamic creation of service interfaces and the transmission of network object references in invocations. We allow the dynamic types of network object references to propagate between separate programs. Finally we provide automatic, fault-tolerant and efficient distributed garbage collection of network objects. In each case, we discuss the requirements of a useful design and the tradeoffs necessary in a real implementation. Our implementation runs on stock systems connected by standard local and wide area networks and internetworking protocols. We believe our approach would support additional library-level tools for security, stable storage, distributed transactions and transparent service replication, though we have not pursued this.

The dissertation demonstrates that it is practical to retain many important amenities of modern programming languages when providing support for the construction of applications in a heterogeneous and evolving distributed system.

Full text

PDF (9.1 MB)

BibTeX record

@TechReport{UCAM-CL-TR-332,
  author =	 {Evers, David Martin},
  title = 	 {{Distributed computing with objects}},
  year = 	 1994,
  month = 	 mar,
  url = 	 {https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-332.pdf},
  institution =  {University of Cambridge, Computer Laboratory},
  doi = 	 {10.48456/tr-332},
  number = 	 {UCAM-CL-TR-332}
}