Using the Linda distribution paradigm to address a variety of computational balancing concerns

Abstract

The design and efficient operation of a distributed information system to support scientific modeling is subject to a variety of balancing concerns, such as conlputation, data, and communication) load and constraints on the use of restricted-license third-party software conlpoIlents, special purpose 1/0 devices, and large data resources. The Regional Ecosystem Information System (REIS described here uses object-oriented principles to separate abstract system objects from site-specific, representation dependeut data files and executable. REIS is implemcmtcd using an object, manager that provides the translation between abstract accesses and accesses to physically distributed system elements, and a distribution manager based on a version of the Linda distribution paradigm that supports actual access to physically distributed objects. Au overview of REIS design is presented, along with preliminary results on some basic performance attributes of this approach to distribution and balancing. OverviewThe design of a modern networked inforlnation system requires consideration of a number of conflicting factors related to system performance, cost, and configuration. C,enerally the system designer must "balance" these factors to determine how a configuration can be built to provide acceptable performance at a reasonable price. There is extensive computing literature related to balancing compution load and message traffic in a distributed system in the execution of a particular algorithm or "nlix" of processes. However, a system designer typically has to address much more severe a pr-Jor~con-figuration constraints before the system evcu begins to Permission to copy without fee all or part of this material ia granted provided that the copies are not made or distributed for direct commercial edvantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying ia by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. Q 1992 ACM O-89791 -502 -X/92 /0002 /0767... S 1.50 function, such as what resources will exist, where they will be located, and how they can be used before tlIe traditional balancing approactles can even I)e corlsidered.Our application is one with these sort of complex configuration balancing factors. It involves potentially lengthy computations, large cumulative data storage requirements and large individllal data objects, interfaces with expeusivr specialized input/output devices. and interfaces with expensive third-party software subsystems, [n such an ellvironmerrt the rr prlon" /JalanCln~de-cisions inhere)]t i]] the selm-t ion of particular peripheral configurations, mapping of (Iata ot~jects to physical data storage, an(i software licensing arrangement is likely to have much more impact 011 system pcrforrtlanre than any subsequent attempt to (Distribute t be rolllputatiolõ r message loa~l across tile n,sourcrs. Giv(>n the importance of these configuration dt'cisions and t...