Work

Organization of Computations in Conditions of Limited Amount of Content Addressable Memory in the Parallel Dataflow Computing System "Buran"

Year: 2019

Type: article

Abstract: The parallel dataflow computing system "Buran" implements the dataflow computing model with a dynamically formed context.To ensure effective parallelization of tasks, it is necessary to organize the operation of the system in conditions of a limited amount of content addressable memory of keys, which is part of the computational core.The article describes the composition and operation of the parallel dataflow computing system, problems associated with the specifics of the computation organization in this system, as well as methods of computation organization to eliminate the possibility of overflow of the content addressable memory of keys the computational core of the system.Such methods are: the computation organization method using a data input block; the method using token-"lock"; the method using division of tokens into stages and swapping out "passive" stages into external storage; the method using "postponed" stages only.These methods allow the solution of tasks with a large amount of data, by having a limited amount of the content addressable memory in the parallel dataflow computing system.

Source: Problems of advanced micro- and nanoelectronic systems development

DOI: 10.31114/2078-7707-2019-2-52-57

Authors: N.N. Levchenko, A.S. Okunev, D.N. Zmejev, A.V. Klimov, A.L. Stempkovsky

Institution: Institute for Design Problems in Microelectronics

Language: English

Topic: Parallel Computing and Optimization Techniques

Subfield: Hardware and Architecture

Field: Computer Science

Domain: Physical Sciences

Open Access status: bronze

Funder: Russian Foundation for Basic Research

Award: G5415934422