V.B. Silin, V.V. Malygin BASIC APPROACH TOWARDS DISTRIBUTING OF COMPUTATION AND INFORMATION RESOURCES IN COMPUTING NETWORKS
Great demand to the quality of computer aided design (CAD) states a problem regarding the quality of the CAD systems themselves. Specificity of such design task causes additional specification and detailed elaboration of all its stages depending on the results of its functioning. Design of CAD system implies the definition not only of its hardware and software structure but also the organization of its operation. It means that static distribution of computation capacity and connections between computing system elements must be defined as well as dynamic of information transfer process must be investigated. The last means that CAD system itself must be considered as a computing network.
Having used body of mathematics along with common obligatory requirements to the function decomposition additional requirements were formed. New requirements represent the fact that the design process should be realized by using some predefined computer type and local area network that brings all particular computers into one information-calculating system. The major of them, from one hand, is to get functionally tied-up and completed particular functions (PF) of the design process and estimation of it for needed computation power, memory requirements for input, output, executable and service information and from another hand, to match these estimations with engineering feasibility on current resources. If the last requirement is not fulfilled then function decomposition must be continued until it meets all the requirements.
After the acceptance of result of function decomposition of primary design task it is possible to calculate required power of computation process for solving all PFs within design time by simple sum of the corresponding figures which were discovered during functional decomposition stage. Calculating of the internal traffic for the design time makes similar estimation for the computing network. In this case total amount information on input and output side of PF is evaluated as well as one that must be input and output from the whole system. Such calculation is performed by processing of information flow diagram and assign weight to every network link corresponding to the value of information flow going through it.
After the estimations mentioned above that give necessary condition of design task solution feasibility using existing system are fulfilled the algorithm for the choice of the best variant of design task realization is proposed. It distributes PFs among computers and saying nothing about design time estimates two major criteria. The first is completeness of the computation power utilization (existed or required). The second is system network links capacity and completeness of the its use. The estimation itself is done by calculating either integrated criterion uses as an input all criterions called above or optimization of network capacity and completeness of its use while other criteria used as the constraints.
Prior to begin finding the best variant of design task realization, a cardinality of combinative space of the feasible variants is to be estimated. It can be done by analyzing two square matrixes of Possible Transitions does it. The first one has dimension equal to number of PF and is used for calculation the number of variants because of variation of PF allocated in the particular nodes with each other. The second one has dimension equal to number of network nodes and is used for calculation the number of variants because of variation groups of PF allocated in nodes with PF groups are allocated in other nodes.
Basing on method of directional transitions (so called “branches and bounds method”) the method of finding quasi-optimal variants of PF distribution was worked out. So together with PF Per Node Distribution Matrix the Information Flows Per Network Links Distribution Matrix is formed on every steps. From formal point of view it is performed by choosing PFs that could be fulfilled on the next step (PF Execution Sequence Matrix is used) then using Information Flow Matrix all preceding PFs (means at least one output data parameter from preceding PF is used as a input one in successor PF) are found out. Having PF coherence information and PF Per Node Distribution Matrix node coherence and final link load are discovered. Then every link is assigned a weight in accordance with the link load and a penalty if the load is higher predefined threshold set. In order to calculate the Step Cost all network link weights are added together. The sum of Step Cost with cost of all preceding steps forms the Cost of PF Distribution (partial or complete). The method result is a PF distribution variant that has minimal Distribution Cost. Also the Integral Quality Criterion is calculated in order to explore how close the found variants are to theoretical maximum of computing system resources utilization.
By this step first static stage of the defined task solution is finished while dynamic CAD functioning optimization is not totally completed and is to be investigated later on.
The new step of automation of manufacture - virtual manufacture (VM) for one of the industrial enterprises is considered (examined), at the expense of integration CAE/CAD/CAM/PPS on the basis of CALS-technology and association of stages of life cycle of a product (LCP) by uniform information model with the automated parallel and three-level management. At the first level the methods of pure(clean) artificial intelligence are used, since informal ways of decomposition of initial model of a product and formation of stages LCP here are applied. At the second level - level of one of stages LCP, which is certain(determined) at the first level, it is expedient to apply the mixed methods of artificial intelligence and CAD. At the third level - level of "details" the methods based, basically, on heuristics are applied pure CAD.
In work the method of search of controlled components both structures of manufacture, and products directed on maintenance of high quality to let out production, based on displacement of support LCP on stages designing - operation is offered. The method allows to refuse from сyrcle at the expense of decomposition of initial model of a product on stages with minimization of communications (connections) between them and to proceed (pass) to parallel designing and manufacturing of a product, as is based on use of the approach as "from above - downwards", where general common model at first is under construction, and "from below - upwards", making a start from known of technologes of manufacturing of details, which are organized on manufacture (CAM), creating units, blocks etc. down to general(common) model, which is concretized and is specified. This conflict is authorized if to use them both, but always it is necessary to begin from details (from below), so that with the subsequent work (from above) organizing concept of the expert laying in model a products, was well supported with the information and that the details of the given development were entered in not without efforts.