Practical paradigm for parallel processing problems

Technical Report · Sat Mar 01 04:00:00 EST 1986

OSTI ID:6007678

Davidson, G S

Data flow computers achieve high performance by keeping as many processing elements busy as possible for a given algorithm. However, without specialized architectures, the synchronization overhead of data-driven computations can prevent significant speedup. To address this problem, the concept of task flow is developed for more general purpose multiprocessors. Task flow differs from data flow in the amount of computation between synchronizations (at the expense of some parallelism). A software technique is described that implements task flow for general purpose multiprocessors with local and shared memories. A SANDAC-IV multiprocessor example of this technique is given, together with timing results. 8 refs., 10 figs., 1 tab.

• Research Organization: Sandia National Labs., Albuquerque, NM (USA)
• DOE Contract Number: AC04-76DP00789
• OSTI ID: 6007678
• Report Number(s): SAND-85-2389; ON: DE86009274
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
ALGORITHMS
ARCHITECTURE
ARRAY PROCESSORS
COMPUTERS
DATA PROCESSING
DATA-FLOW PROCESSING
ELECTRONIC CIRCUITS
MATHEMATICAL LOGIC
MICROELECTRONIC CIRCUITS
MICROPROCESSORS
PARALLEL PROCESSING
PROCESSING
PROGRAMMING
SYNCHRONIZATION
TASK SCHEDULING