Interested in reading more?

A new form of computation is emerging. Propelled by advances in software design and increasing connectivity, networks of enormous complexity known as distributed computational systems are spreading throughout offices and laboratories, across countries and continents. Unlike standalone computers, these growing networks seldom offer centralized scheduling and resource allocation. Instead, computational processes (the active execution of programs) migrate from workstations to printers, servers, and other machines in a network as the need arises, without a priori knowledge of the state or availability of other network components at run time. This novel form of delocalized and asynchronous computation is called open because the closed-world assump tion of traditional computer science is no longer valid. In a closed system, the computer executes its programs at the command of a central scheduler that has global knowledge of the system. But in an open system, processes make local decisions on how to access given resources despite imperfect knowledge about the system and information that is at times inconsistent and delayed. In short, processes in an open system can behave very much like agents in a biological community. The appearance of such complex systems on the computational scene creates a number of challenging design and implementation problems for computer scientists. Processes in an open system lack a global perspective because they cannot know what other kinds of processes will be interacting with their own—much less how they’ll behave. Therefore, what is required is a whole different approach to system-level programming and the creation of new languages. Biological organizations have evolved to operate successfully despite imperfect knowledge of their environment, but it is far from obvious how one would mimic those characteristics in an open computational system. Nevertheless, a serious effort at designing such systems is already underway at a number of laboratories. Some of these prototype systems can, in fact, adapt to unforeseen conditions. For example, the Enterprise system, implemented at Xerox PARC and MIT, is a market-like scheduler in which independent processes or agents are allocated at run time among remote idle workstations through a bidding mechanism. This system offers substantial improvements over more conventional systems, even when it encounters large delays and lacks accurate estimates of processing times. This brings to mind the spontaneous appearance of organized behavior in biological and social systems. Biological agents, like bees or ant, engage in cooperating strategies while working on the solution of particular problems. in some cases, their behavior moves toward an asymptotic state, one that is immune to small perturbations. This phenomenon sometimes goes under the name of evolutionarily stable strategy (ESS) and is being intensely studied by biologists and social scientists. Recently, this kind of asymptotic behavior has been observed in open computational systems as well. Computational agents, programs—in experimental open systems exhibit spontaneous organization. That is, the programs work cooperatively while achieving their individual goals. However, imperfect knowledge of the global network’s functions and delays in processing information introduce chaos, thereby preventing the systems from achieving an evolutionarily stable state. While at the moment they operate less than perfectly, these computational ecosystems are posing fascinating questions concerning their function, dynamics, and efficiency. For example, scientists would like to understand, and even predict, overall system behavior from knowledge of what the individual processes can do. Since a system designer can arbitrarily set the rules that determine how computational agents choose among possible strategies for accomplishing their tasks, whole scenarios could be created and tested. Moreover, scientists can gain insights from the similarities between computational and biological open systems. Just as insect colonies exhibit the workings of a natural computational system, the dynamics of an artificial open system offer quantitative testing grounds for models of social organizations that are hard to design in natural settings. Open systems is a new field and the pace of its growth is swift. As very large computer communications systems such as Bitnet and Arpanet are used for accessing remote supercomputers and special-purpose facilities, distributed computation becomes increasingly widespread and more complex. At the same time, new vistas in its understanding are appearing through the use of nonlinear game dynamics, market models, and high-level concurrent languages. We are witnessing the spread of a computational membrane across vast areas of the Earth, and this is radically changing the way we use and think about computers. Equally important, it is leading to a new field of interdisciplinary research rich in phenomena and deep in structure. B.A. Huberman is a faculty member of the Department of Applied Physics, Stanford University, and is a Research Fellow at the Xerox Palo Alto Research [Ed. note: For further reading, see The Ecology of Computation, B. A. Huberman, editor. North Holland, Center, Palo Alto, Calif. 1988.]

The Ecology of Computation

May 15, 1988

Interested in reading more?

Become a Member of