| Implementation of Majority Function Based on Matching Processor in the Parallel Dataflow Computing System "Buran" |
|
 |
|
|
| Authors |
| | Levchenko N.N. |
| | Zmejev D.N. |
| Date of publication |
| | 2022 |
| DOI |
| | 10.31114/2078-7707-2022-3-129-134 |
|
| Abstract |
| | The reliability is one of the problems facing the developers of supercomputers with tens and hundreds of thousands of computational cores. Systems based on non-traditional com-puting models must also provide the required level of relia-bility. One of such systems is a promising dataflow compu-ting model with a dynamically formed context. This computing model is implemented by the parallel dataflow compu-ting system (PDCS) "Buran", to which the parallelism of computations immanently inherent at the hardware level. One of the main methods for improving reliability is modular redundancy. This method fits well with the decentralized PDCS architecture and the distribution of computations supported at the hardware level. The dataflow computing model with a dynamically formed context is characterized by the activation of computational quanta by data readiness. A computational quantum is a relatively small program that is activated as soon as all the data is received at its input. These quanta interact with each other using messages - data structures containing an operand, service fields and context. The context uniquely defines the position of the operand in the virtual address space of the task. Due to the specifics of the operation of the matching processor and the content-addressable memory included in it, the majority function (modular redundancy) can be implemented directly in the matching processor without additional hardware costs. The implementation of the majority function using the matching processor is possible in the following ways: use of context fields, use of standard tokens, use of tMJR special tokens and use of multi-input tokens. The article discusses these options, the advantages and disadvantages are given |
| Keywords |
| | parallel dataflow computing system, majority function, matching processor, content-addressable memory |
| Library reference |
| | Levchenko N.N., Zmejev D.N. Implementation of Majority Function Based on Matching Processor in the Parallel Dataflow Computing System "Buran" // Problems of Perspective Micro- and Nanoelectronic Systems Development - 2022. Issue 3. P. 129-134. doi:10.31114/2078-7707-2022-3-129-134 |
| URL of paper |
| | http://www.mes-conference.ru/data/year2022/pdf/D023.pdf |
