Data Processing under the Control of Descriptors
in Parallel Computing Systems

Valeriy Zhabin

Computer Engineering Department, National Technical University of Ukraine
“Kiev Polytechnical Institute”, Peremogy pr., 37, Kiev, 03056, Ukraine, e‑mail: vizhabin@comsys.ntu-kpi.kiev.ua

The concept of automatic dynamic distribution of jobs on computing modules in parallel systems is considered. For parallel programs developing a different programming tools are created. Besides the parallel languages, which are independent means for programming, different extensions of sequential languages, as well as libraries of parallel programming find their use. The most of known technologies refer to means of static process paralleling. In this case the main problems of paralleling are resolved by the programmer on the programming stage. The common imperfection of static paralleling is that during the static analysis of algorithms it is not always possible to reveal parallel branches. It is due to many reasons, the basic of which is the lack of information about dynamic of processes.

The methods of automatic dynamic paralleling of algorithms with a coarse-grained structure under the control of descriptors of the data and jobs in systems with the conventional architecture are offered. The offered method of implementation of calculations in parallel systems allows to eliminate a set of problems, bound with conventional static planning. The process of preparation of a task is essentially simplified. There is no necessity to take into account the duration of execution of jobs, to reveal parallel branches, to distribute data on the basis of the static analysis. The mandatory registration of all tasks before the beginning of the score is not required, that allows starting the solution of a new task at any moment, irrespective of a conditions of other tasks. The preparation of tasks does not depend on number of computing modules in a system. Due to this the system reconfiguration does not result in necessity of resolving a task. Besides the solution of a problem of maintenance of systems liveness is considerably simplified.