Authors:
  K. Mani Chandy, Boris Dimitrov, Huy Le, Jacob Mandelson, Matthew Richardson,
  Adam Rifkin, Paolo A.G. Sivilotti, Wesley Tanaka, and Luke Weisman

Affiliation:
  California Institute of Technology

Email Address:
  adam@cs.caltech.edu

A WORLD-WIDE DISTRIBUTED SYSTEM USING JAVA OBJECTS AND THE INTERNET
-------------------------------------------------------------------

OVERVIEW

This paper describes the design of a distributed system built using Java
that supports peer-to-peer communication among processes spread across a
network.  We identify the requirements of a software layer that supports
distributed computing, and we propose a design that meets those
requirements.  Our primary concern is the identification, specification,
and implementation of software components that can be composed in
different ways to develop correct distributed applications.  Though our
implementation uses Java, the fundamental ideas apply to any
object-oriented language that supports messaging and threads.

MOTIVATION

Millions use the World-Wide Web for information exchange and for
client-server applications. Widespread use has led to the development of
a cottage industry for producing Web-based documentation; large numbers
of people without formal education in computing are developing server
applications on the Web. This paper describes a project to extend this
infrastructure to a distributed system with peer-to-peer process
communication across the Global Information Infrastructure (GII).
The focus of our project is on identifying and specifying software
components that can be composed to create distributed applications,
implementing the software components as classes in an object-oriented
framework, developing a compositional methodology for constructing
correct distributed applications from the components, and implementing a
library of applications that demonstrates the methodology.

Our project develops methods and tools for distributed programming
applications layered on top of standard network technologies. Though our
implementation uses Java, we could have used any other object-oriented
language that supports threads and communication classes.

CORRECTNESS 

Our project deals with providing software components and compositional
methods that support the development of correct distributed
applications.  The methods employed by Web users for developing
client-server applications are not the best methods for developing
correct peer-to-peer distributed applications.  Furthermore, approaches
for debugging sequential programs are inadequate for ensuring
correctness in distributed applications.  Our challenge is to deal with
the difficult problems of distributed systems --- problems such as
deadlock, livelock, and sending unbounded numbers of messages --- that
are not issues in sequential programs.

CONTRAST WITH TRADITIONAL DISTRIBUTED SYSTEMS

Certain kinds of distributed systems are inherently complex, deal with
myriad functions, have strict performance requirements, and have
disastrous consequences of failure; examples of these systems include
applications in telemedicine, air traffic control, and military command.
Such systems are developed in a painstaking manner, with each system
design led by a single group of expert designers that has primary
responsibility for the entire system.  By contrast, many Web-based
applications are relatively simple, are collaborative by nature, have
limited functionality, are performance-limited by the GII, and may be
developed by people who are not experts in concurrent computing;
examples of these systems are a calendar scheduling application and
collaborative interactive environments.  Designers of such applications
have little control over the networks, protocols, operating systems, and
computers on which their applications execute.

Research issues for the two classes of distributed systems are somewhat
different.  The former class of distributed systems, the class on which
human lives depend, is extremely important and has benefited from a
great deal of research.  This paper, however, deals with facilitating
the development of the latter class of applications.  The class of
collaborative Web-based applications, though less critical than
traditional distributed applications, has interesting engineering
challenges nonetheless.

DAPPLETS AND SESSIONS

We coin the phrase dapplet to distinguish a process used in a
collaborative distributed application from processes used in traditional
distributed systems.  Dapplets are composed together to form distributed
sessions.  A session is a temporary network of dapplets that carries out
a task such as arranging a meeting time for a group of people.  Sessions
need not be static; after initiation, they may grow and shrink as
required by the dapplets.

A ten-page position paper describing dapplets and sessions is
available on the World-Wide Web

    Postscript version:
       http://www.cs.caltech.edu/~adam/CALTECH/papers/chandy_etal.ps

    HTML version:
       http://www.cs.caltech.edu/~adam/CALTECH/papers/chandy_etal/

The overall EDEN (Experimental Distributed ENvironments) home page
on the Web is available at

    http://www.cs.caltech.edu/~adam/CALTECH/eden.html