Distributed computing is the most natural environment for the MOVIE System. The communication model for MOVIE networks is based on one simple principle, uniform for distributed and MIMD-parallel architectures: nodes of such network communicate by sending MovieScript. This model unifies communication and computation: Computing in MOVIE is when a server interprets MovieScript, whereas communication occurs when a server sends MovieScript to be interpreted by another server on the network.
Social human activities provide adequate analogies here. One can think of MOVIE network as a society of autonomous intelligent agents, capable of internal information processing and of information exchange, both organized in terms of the same high-level language structures. The processing capabilities of such a system are in principle unlimited. Detailed programming paradigms for distributed computing are not specified initially at the MovieScript level and can be freely selected depending on the application needs. Successful computation/communication patterns with some reusability potential can then be retained within the system in the form of appropriate MovieScript extensions.
The MovieScript-based user-level model for MOVIE networks is uniform, elegant, and appealing. Its detailed technical implementation, however, is a complex task. Communication services must be included at the lowermost level of the inner loop of the MovieScript interpreter and coordinated with scheduling, event handling, software interrupts, and other dynamic components of the server. Also, when building interfaces to existing Open Systems components, we have to cope with various existing network and message-passing protocols. In the networking domain, we use the UNIX socket library as the base-C-level platform, but then the question arises of how to handle higher protocols such as NFS, RPC, XDR and a variety of recent ``open'' models (see, e.g. Figure 17.4). Similar uncertainties arise in handling and integrating various message-passing protocols for MIMD-parallel computing.
Figure 17.4: An Example of the Distributed MOVIE Environment. The figure
illustrates various network-extensible graphics protocols used in
implementing the uniform high-level MovieScript protocol. We denote by
mps, nps, and dps, respectively, the MOVIE, NeWS, and
Display PostScript protocols. MOVIE servers communicate directly via
mps, whereas MovieScript messages sent to NeWS/DPS/X servers are
internally translated by the MOVIE server to the remote server-specific
protocols.
Since the consistent implementation of the MovieScript-based communication is one of the most complex tasks in the MOVIE development process, we adopted the following evolutionary and self-supporting approach. The system design was started in the single-node, single-thread configuration. The notion of multithreading was built into the design from the beginning by adopting consistent thread-relative addressing modes. In consequence, the detailed model for scheduling, networking, and message passing factorized as an independent sector of MovieScript and it was initially postponed. The base interpreter loop was developed first. In the next stage, we constructed the field algebra for regular matrix processing, interpreted object-oriented model with rapid prototyping capabilities and graphics/visualization/windowing layers with the focus on interpreted GUI interfaces.
These layers are currently in the mature stage and they can now be used to provide GUI support for prototyping multithreading distributed MOVIE networks, starting with the regular modules for concurrent matrix algebra and signal processing. The current status of the design and implementation work on scheduling, networking and message passing is described in [Niemiec:92a], [Niemiec:92b] and [Furmanski:93a].