structures for general purpose parallel processing.
CCA is also appropriate for other technologies that
exhibit high W values due to very high speed clock
rates, such as superconductor RSFQ technologies.
Unlike conventional cellular automata, CCA
incorporates mechanisms in support of the semantics
of a general and global model of parallel computing.
The operation of the CCA highly parallel system is
governed by the ParalleX computing model
developed under the leadership of Argonne National
Laboratory in collaboration with the University of
Delaware and managed by an ultra lightweight kernel
runtime software system being developed in
collaboration with Sandia National Laboratory and
the University of New Mexico. The computing model
reflects a message-driven split-transaction processing
discipline for efficient latency tolerant fine-grain
parallel processing and relies on a combination of
dataflow and futures based synchronization.
CCA may provide the ultimate convergent
architecture for nanoscale and ultra high clock rate
technologies at the end of Moore’s Law. In the limit,
it constrains the local computing element to the
minimum size capable of performing a single
operation through the merging of memory, logic, and
communication hardware in a single component.
Therefore, it will support the smallest possible
feature size and highest clock rate possible. For even
higher W, the CCA cell can be pipelined and handle
overlapping incident request packets.
structures for general purpose parallel processing.
CCA is also appropriate for other technologies that
exhibit high W values due to very high speed clock
rates, such as superconductor RSFQ technologies.
Unlike conventional cellular automata, CCA
incorporates mechanisms in support of the semantics
of a general and global model of parallel computing.
The operation of the CCA highly parallel system is
governed by the ParalleX computing model
developed under the leadership of Argonne National
Laboratory in collaboration with the University of
Delaware and managed by an ultra lightweight kernel
runtime software system being developed in
collaboration with Sandia National Laboratory and
the University of New Mexico. The computing model
reflects a message-driven split-transaction processing
discipline for efficient latency tolerant fine-grain
parallel processing and relies on a combination of
dataflow and futures based synchronization.
CCA may provide the ultimate convergent
architecture for nanoscale and ultra high clock rate
technologies at the end of Moore’s Law. In the limit,
it constrains the local computing element to the
minimum size capable of performing a single
operation through the merging of memory, logic, and
communication hardware in a single component.
Therefore, it will support the smallest possible
feature size and highest clock rate possible. For even
higher W, the CCA cell can be pipelined and handle
overlapping incident request packets.
การแปล กรุณารอสักครู่..
